Effects of resistance training performed to repetition non-failure on exercise performance in healthy adults: a systematic review and meta-analysis

Abstract

Objective

This systematic review and meta-analysis aimed to compare the effects of resistance training (RT) performed to non-failure versus failure on muscular strength, hypertrophy, endurance, and power in healthy adults.

Method

We searched four databases (PubMed, Web of Science, SPORTDiscus, and MEDLINE). Randomized controlled trials comparing RT performed to non-failure and failure were included. Outcomes included muscular strength, hypertrophy, endurance, and power. Random-effects meta-analyses were performed to calculate standardized mean differences (SMD) with 95% confidence intervals (CI).

Result

A total of 20 studies with 556 participants were included. The results indicated non-failure training superior to failure training in terms of dynamic strength (SMD_pooled = 0.24, 95% CI: 0.06 to 0.42, p = 0.01). However, no significant effects were observed for isometric strength, muscular hypertrophy, muscular endurance, and muscular power. Subgroup analysis revealed that participant characteristics, training volume, training method, and training duration were moderating variables influencing training effects.

Conclusion

These findings indicate that resistance training performed to non-failure appears to be as effective as training to failure for most neuromuscular adaptations and may provide a small advantage for dynamic strength development. These findings suggest that reaching muscular failure is not necessary to maximize resistance training adaptations. Future research should explore the long-term physiological and neuromuscular adaptations associated with non-failure training across different populations, training statuses, and sport events.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13102-026-01861-z.

Introduction

Resistance Training (RT) is recognized as an effective strategy for improving muscle function and exercise performance [1–3]. Previous studies have shown that RT can enhance muscle strength, muscular endurance, muscular power, and sprint performance [4–6]. However, these effects are highly dependent on the manipulation of training variables, including intensity, volume, intervals, and duration [7]. For example, Kraemer et al. reported that the effect of RT mainly depends on training variables such as intensity. Among these variables, training intensity is considered a critical determinant of neuromuscular adaptations [8]. Traditionally, intensity is prescribed either as a percentage of one-repetition maximum (1RM) or based on the number of repetitions performed before muscular failure [9, 10].

Muscle failure is the inability to complete the concentric phase of a repetition against a certain resistance without altering movement technique or the number of repetitions [11]. Training to muscle failure has been widely implemented in RT programs and is often considered a strategy to maximize motor unit recruitment, mechanical tension, and anabolic signaling [12–14]. These mechanisms are believed to promote neuromuscular adaptations that may enhance strength and hypertrophy. However, recent studies indicate that training to failure substantially increases metabolic stress and neuromuscular fatigue, which may impair subsequent force production and training quality [15–19]. Refalo et al. reported that RT performed to failure increased neuromuscular fatigue and reduced movement velocity compared to non-failure RT. Repeated training to failure, which causes excessive fatigue, can increase the risk of overreaching, injury, and longer recovery times.

In contrast, non-failure training has emerged as a strategy that may provide sufficient mechanical and neuromuscular stimulus while minimizing excessive fatigue accumulation [12, 14, 20, 21]. By avoiding failure, athletes may maintain higher-quality repetitions and movement velocity, potentially facilitate recovery, and improve overall training sustainability. Moreover, non-failure training may reduce injury risk and enhance adherence to training programs [22, 23]. However, studies directly comparing RT performed to failure versus non-failure have produced inconsistent findings. Some studies have reported superior adaptations with failure training. For example, Drinkwater et al. reported that training to failure significantly improved strength in elite athletes compared with non-failure training [12]. Conversely, other studies have suggested that training to failure may not provide additional benefits. Folland et al. reported that fatigue is not a necessary stimulus for strength gains during RT [24]. Likewise, Martorelli et al. found that RT performed to failure did not lead to greater improvements in strength or hypertrophy compared with non-failure training [25]. These differences may stem from the manipulation of training variables such as intensity, volume, intervals, and duration, as well as from methodological limitations, including small sample sizes. Furthermore, as RT increasingly performed velocity-based training (VBT), the impact on the differences between failure and non-failure training remains uncertain.

From a mechanistic perspective, training to failure may increase motor unit recruitment and metabolic stress during a set, which could theoretically enhance hypertrophic and strength-related adaptations [26, 27]. However, repeatedly reaching failure also increases peripheral and central fatigue, reduces movement velocity, and may impair technical quality and subsequent training performance [28, 29]. In contrast, non-failure training may provide sufficient mechanical and neural stimulus while limiting excessive fatigue accumulation [30, 31]. This may be particularly relevant for dynamic strength and power outcomes, where movement quality, intermuscular coordination, and the ability to express force rapidly are important [32–35]. Therefore, the expected effects of failure versus non-failure training may be outcome-specific rather than uniform across strength, hypertrophy, endurance, and power. Although previous meta-analyses have examined resistance training performed to failure versus non-failure, most have primarily focused on maximal strength and muscular hypertrophy [30, 36–38]. Less attention has been given to broader exercise performance outcomes, such as muscular endurance and muscular power, which are highly relevant to sport performance and resistance training prescription. In addition, it remains unclear whether participant characteristics, training volume, intervention duration, and velocity-based approaches modify the comparative effects of failure and non-failure training.

Therefore, this systematic review and meta-analysis aimed to evaluate the effects of RT performed on non-failure versus failure on exercise performance. We examined dynamic and isometric strength, muscular hypertrophy, endurance, and power. We also explored whether participant characteristics, training volume, training method, and intervention duration moderated these effects.

Methods

A systematic review and meta-analysis of the literature was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [39]. This study was registered with PROSPERO (CRD42022382889).

Search strategy

Two independent reviewers conducted a comprehensive literature search across four databases (PubMed, Web of Science, SPORTDiscus, and MEDLINE) from inception to 14 December 2025. The detail search strategy was (“resistance training” OR “resistance exercise” OR “strength training” OR “strength exercise” OR “weight training” OR “weight exercise” OR “velocity based training” OR “velocity based” OR “vbt” OR “concentric velocity” OR “mean concentric velocity” OR “movement velocity” OR “barbell velocity” OR “velocity loss” OR “power based training”) AND (“non-failure training” OR “non failure training” OR “repetition failure” OR “repetition exhaustion” OR “failure training” OR “muscular failure” OR “muscle failure” OR “not to failure” OR “to failure” OR “volitional interruption” OR “without resting” OR “low fatigue” OR “high fatigue”) AND (“1 repetition maximum” OR “one repetition maximum” OR “1 RM” OR “1RM” OR “MVC” OR “maximal voluntary contraction” OR “muscle strength” OR “muscular strength” OR “muscle hypertrophy” OR “muscular hypertrophy” OR “muscle fibre” OR “muscle fiber” OR “muscle thickness” OR “CSA” OR “cross-sectional area” OR “muscle size”).

Data sources and study selection

Studies were included if: (1) healthy participants were ≥ 18 years; (2) the intervention involved RT performed to non-failure; (3) the comparison involved RT performed to failure; (4) Outcomes included at least one of the following: muscular strength, muscular hypertrophy, muscular endurance, or muscular power; (5) The study design was a randomized controlled trial.

Studies were excluded if they were: (1) animal trials; (2) written in a non-English language, (3) without specific data; (4) review and conference articles.

Data extraction, outcomes, and risk of bias assessment

Two independent reviewers extracted relevant data from each included study. One reviewer performed the primary extraction, and a second reviewer verified all extracted values against the original full-text articles [40], we extracted the following data: (a) lead author and year of publication; (b) sample size and participant characteristics, including age, body weight, and resistance training levels; (c) details of the resistance training programs (intensity, volume, duration, and frequency); (d) exercise mode (smith machine or free weight); (e) training mode (varied or unvaried intensity); (f) outcome measures. Disagreements were resolved through discussion between the two reviewers. If they couldn’t reach a consensus, a third reviewer made the final decision. The mean and standard deviation for each outcome in each study’s post-tests were extracted. The post-test outcome data for each study was summarized in Table 1.

Table 1.

Characteristics of the included studies (N = 20)

Study	Participant Characteristics	Study design	Simple Size	Mean Age	Non-failure Programs	Failure Programs	Frequency Period	Outcome Measures
Bergamasco et al. (2020) [41]	untrained	RCT	E:12 C1:13 C2:13	Male:67 ± 6 Female:64 ± 3	40% 1RM leg press, leg curl: 3 sets * reps to voluntary interruption	40% 1RM leg press, leg curl: (3 sets * 10 reps)	2/week 12week	Dynamic strength: 1RM leg press↑ Muscular hypertrophy: Vastus lateralis CSA↑ Muscular power: Rate of torque development↑ Others: Chair stand↑ Habitual gait speed↑ Maximal gait speed and Timed up-and-go remained unchanged
Colomer-Poveda et al. (2021) [42]	untrained	RCT	E:11 C1:11 C2:11 C3:9	E1:21 ± 1.3 C1:21 ± 1.4 C2:22 ± 1.5 C3:24 ± 4.3	75% 1RM knee extension: (6 sets * 5/6 reps)	25% 1RM knee extension: 3 sets * reps to failure 75% 1RM knee extension:3 sets * reps to failure	4/week 4week	Dynamic strength: 1RM knee extension: Failure(high load)↑20% Non-failure(high load)↑15% Failure(low load) and Control remained unchanged Isometric strength: Maximal voluntary isometric force: Failure(high load)↑6% Non-failure(high load)↑12% Failure(low load) and Control remained unchanged Others: Failure > Non-failure: RPE Failure = Non-failure: VL-EMGRMS remained unchanged, RF-EMGRMS↑ Tw-Mmax:Failure(high load)↓-10% Non-failure(high load)↓-0.5% Failure(low load)↓-19% Control↓-2%
Dorrell et al.(2020) [43]	trained	RCT	E + C:16	22.8 ± 4.5	70–90% 1RM back squat: 3 sets * programmed reps 70–92% 1RM bench press :3 sets * programmed reps	70–90% 1RM back squat: 3 sets * programmed reps 70–92% 1RM bench press: 3 sets * programmed reps	2/week 6weeks	Dynamic strength: 1RM squat: Failure↑8%,Non-failure:↑9% 1RM bench press: Failure↑4%< Non-failure:↑8% 1RM strict overhead press: Failure↑6%,Non-failure:↑6% 1RM deadlift: Non-failure:↑6%,Failure remained unchanged Muscular power: CMJ: Non-failure↑5% Failure remained unchanged
Drinkwater et al.(2005) [12]	trained	RCT	E + C:26	B-er:18.6 ± 0.3 F-er:17.4 ± 0.5	80–105% 6RM bench press: (8 sets * 3 reps)	80–105% 6RM bench press: (4 sets * 6 reps)	3/week 6week	Dynamic strength: Failure > Non-failure: 6RM bench press↑ Muscular power: bench throw power↑
Folland et al.(2002) [24]	untrained	RCT	E:11 C:12	E:20 ± 1 C:22 ± 2	73% 1RM knee extension: 40 reps with 30 s between each repetition	73% 1RM knee extension: (4 sets * 10 reps)	3/week 9week	Dynamic strength: 1RM knee extension: Failure↑34%, Non-failure↑40% Isometric strength: Failure = Non-failure: isometric knee extension↑ Others: Angle-torque relation↑, torque-velocity relation↑
Held et al.( 2021) [44]	trained	RCT	E + C:21	19.6 ± 2.1	80% 1RM squat, deadlift, bench row, bench press: 4 sets * reps to 10% velocity loss	80% 1RM squat, deadlift, bench row, bench press: 4 sets * reps to failure	2/week 8week	Dynamic strength: Failure < Non-failure: 1RM squat↑, 1RM deadlift↑ 1RM bench row↑, 1RM bench press↑ Others: Failure > Non-failure: overall stress Failure = Non-failure: VO2max↑, PVO2max↑ Failure < Non-failure: overall recovery
Izquierdo et al.(2006) [22]	trained	RCT	E:15 C:14	E:23.9 ± 1.9 C:24.8 ± 2.9	1RM bench press, parallel squat: (3 sets * 2–4 reps)	1RM bench press, parallel squat: (3 sets * 2–4 reps)	2/week 11week	Dynamic strength: 1RM parallel-squat: Failure↑22%, Non-failure↑23% 1RM bench press: Failure↑23%, Non-failure↑23% Muscular endurance: Failure > Non-failure: bench press endurance↑ Failure = Non-failure: parallel-squat endurance↑ Muscular power: Parallel-squat power: Failure↑26%, Non-failure↑29% Bench press power: Failure↑27%, Non-failure↑28% Failure = Non-failure: CMJ↑
Karsten et al.(2021) [45]	trained	RCT	E:9 C:9	E:23 ± 5 C:24 ± 4	75% 1RM bench press, parallel squat: (8 sets * 5 reps)	75% 1RM bench press, parallel squat: 4 sets * 10RM reps to failure	2/week 6week	Dynamic strength: Failure > Non-failure:1RM bench press Failure < Non-failure: 1RM squat Muscular hypertrophy: Failure > Non-failure: vastus medialis muscle thickness Failure < Non-failure: anterior deltoid muscle thickness Elbow flexor muscle thickness: Failure↑, Non-failure remained unchanged Muscular power: Bench press power: Failure remained unchanged, Non-failure↑ Failure = Non-failure: CMJ remained unchanged Others: Failure > Non-failure: fat, fat percentage, Failure = Non-failure: arm and thigh circumference remained unchanged
Kramer et al.(1997) [46]	trained	RCT	E1:14 E2:13 C:16	20.3 ± 1.9	10RM squat, push press, bench press, pull from mid-thigh, leg curl, bent-over row, crunch: (3 sets * 10 reps)	8–12RM squat, push press, bench press, pull from mid-thigh, leg curl, bent-over row, crunch (1 set * 8–12 reps)	3/week 14week	Dynamic strength: Failure > Non-failure: 1RM squat↑ 1RM squat/body mass↑
Lacerda et al.(2020) [47]	untrained	RCT	E + C:10	23.7 ± 4.9	Not reported	Not reported	2–3/week 14week	Dynamic strength: Failure = Non-failure: 1RM knee extension↑ Isometric strength: Failure = Non-failure: knee extension MVIC↑ Muscular endurance: Failure = Non-failure: knee extnsion MNR↑ Muscular hypertrophy: Failure = Non-failure: rectus femoris and vastus lateralis CSA↑ Others: Failure = Non-failure: EMGRMS remained unchanged
Lasevicius et al. (2019) [48]	untrained	RCT	E + C:25	high:23.8 ± 4.9 low:24.3 ± 4.8	80% 1RM knee extension (high load): (5.5 sets * 6.7 reps) 30% 1RM knee extension (low load): (5.4 sets * 19.6 reps)	80% 1RM knee extension (high load): (3 sets * 12.4 reps) 30% 1RM knee extension (low load): 3 sets * 34.4 reps)	2/week 8week	Dynamic strength: 1RM knee extension: failure(high)↑ non-failure(high)↑, failure(low)↑ Muscular hypertrophy: High load > Low-load: quadriceps CSA Others: Failure > Non-failure: RPE
Martorelli et al.(2017) [35]	untrained	RCT	E1:32 E2:27 C:30	E1:21.7 ± 2.8 E2:21.6 ± 3.3 C:22.3 ± 3.8	70% 1RM elbow flexor: (4 sets * 7 reps) 70% 1RM elbow flexor: (3 sets * 7 reps)	70% 1RM elbow flexor: 3 sets * reps to failure	2/week 10week	Dynamic strength: Failure = non-failure: 1RM elbow flexor Muscular endurance: Failure = non-failure: elbow flexor muscle endurance↑ Muscular hypertrophy: elbow flexor muscle thickness: Failure and Non-failure (volume equated)↑,Non-failure (volume non-equated) remained unchanged Muscular power: Peak torque at 60°/s: Non-failure (volume equated)↑,Failure and Non-failure (volume non-equated) remained unchanged Peak torque at 180°/s: Non-failure (volume equated) and Non-failure (volume non-equated)↑, Failure remained unchanged
Nóbrega et al.(2018) [49]	untrained	RCT	E + C:27	23.0 ± 3.6	80% 1RM knee extension (high load): 3 sets * reps to volitional interruption 30% 1RM knee extension (low load): 3 sets * reps to volitional interruption	80% 1RM knee extension (high load): 3 sets * reps to failure 30% 1RM knee extension (low load): 3 sets × reps to failure	2/week 12week	Dynamic strength: Failure = non-failure: 1RM knee extension↑ Muscular hypertrophy: Failure = non-failure: vastus lateralis CSA↑, PA↑ Others: High load > Low load: EMG
Pareja-Blanco et al.(2017) [50]	trained	RCT	E:12 C:10	22.7 ± 1.9	70–85% 1RM full (deep) squat: 3sets * reps to 20% velocity loss	Not reported	2/week 8week	Dynamic strength: 1RM squat: Failure↑13.4%, Non-failure↑18.0% Muscular endurance: Failure = Non-failure: T20 remained unchanged, Muscular hypertrophy: Failure > Non-failure: vastus lateralis + vastus intermedius muscle volume Failure = Non-failure: muscle fiber CSA↑, quadriceps muscle volume↑, vastus medialis muscle volume↑, rectus femoris muscle volume remained unchanged Muscular power: Failure < Non-failure: CMJ Others: AV: Failure↑6.0%, Non-failure↑12.5% AV > 1: Failure remained unchanged, Non-failure↑6.2% AV < 1: Failure↑13.7%, Non-failure↑21.7%
Rooney et al.(1994) [14]	untrained	RCT	E:13 C:14	18–35	6RM elbow flexion: 6–10 reps/set	6RM elbow flexion: 6–10 reps/set	3/week 6week	Dynamic strength: Failure > Non-failure: 1RM elbow flexion↑ Isometric strength: Failure = Non-failure: isometric elbow flexion↑ Muscular endurance: Failure > Non-failure: isometric fatigue
Sampson et al.(2016) [51]	untrained	RCT	E1:10 E2:8 C:10	E1:23.7 ± 6.2 E2:24.3 ± 7.0 C:23.4 ± 6.6	85% 1RM elbow flexion–extension (rapid shortening, RS): (4 sets * 4 reps) 85% 1RM elbow flexion–extension (stretch-shortening cycle, SSC): (4 sets * 4 reps)	85% 1RM elbow flexion–extension (control, C): (4 sets * 6 reps)	3/week 12week	Dynamic strength: 1RM elbow flexion: Failure↑30.6%, Non-failure (RS)↑28.6%, Non-failure (SSC)↑32.8% Isometric strength: Elbow flexion MVIC: Failure↑11.6%, Non-failure (RS)↑10.9%, Non-failure (SSC)↑7.1% Muscular hypertrophy: Elbow flexion muscle CSA: Failure↑14.3%, Non-failure (RS)↑12.7%, Non-failure (SSC)↑12.8%
Sanborn et al.(2000) [52]	untrained	RCT	E:8 C:9	18.0–20.0	1RM parallel-squat: 3 sets * reps not to failure	1RM parallel-squat: (1 set* 8–12 reps)	3/week 8week	Dynamic strength: 1RM squat: Failure↑24.2%, Non-failure↑34.7% Muscular power: CMJ: Failure↑0.3%, Non-failure↑11.2%
Santanielo et al.(2020) [53]	trained	RCT	E + C:14	Not reported	1RM leg press, leg extension: sets * reps to voluntary interruption	1RM leg press, leg extension: sets * reps to muscle failure	2/week 10week	Dynamic strength: Failure = Non-failure: 1RM leg press 45°↑, 1RM leg extension↑ Muscular hypertrophy: Failure = Non-failure: vastus lateralis CSA↑, pennation angle↑, fascicle length↑, Others: EMG no significant differences between groups
Terada et al.(2022) [54]	untrained	RCT	E:8 C1:9 C2:10	Not reported	40% 1RM bench press: 3 sets * reps to 20% velocity loss 80% 1RM bench press: (3 sets * 8 reps)	40% 1RM bench press: 3 sets * reps to volitional failure	2/week 8week	Dynamic strength: 1RM bench press: Failure (low load) = Non-failure (low load) < Non-failure (high load) Muscular endurance: Bench press muscle endurance: Failure (low load) = Non-failure (low load) > Non-failure (high load) Muscular hypertrophy: Failure (low load) = Non-failure (low load) = Non-failure (high load): triceps brachii and pectoralis major muscle thickness↑ Muscular power: Failure (low load) = Non-failure (low load) = Non-failure (high load): bench press power↑
Vieira et al.(2019) [55]	trained	RCT	E:8 C:6	E:24.5 ± 1.6 C:25.2 ± 2.2	90% 10RM bench press, leg press: (3 sets * 10 reps)	10RM bench press, leg press: (3 sets * 10 reps)	3/week 8week	Dynamic strength: Failure = Non-failure: 1RM bench and leg press↑ Muscular endurance: Failure = Non-failure:10RM bench press, leg press, seated row, and back squat↑ Failure < Non-failure: estimated repetitions to failure Others: Failure > Non-failure: RPE

1RM one repetition maximum, 6RM six repetition maximum, 10RM ten repetition maximum, AV average velocity, CMJ countermovement jump, CSA cross-sectional area, EMG_RMS root mean square of electromyographic signal, MNR maximum number of repetitions, MVC maximal voluntary, MVIC maximal voluntary isometric contraction, PA pennation angle, P_VO2max peak oxygen uptake, RCT randomized controlled trial, RF rectus femoris, RPE rating of perceived exertion, RS rapid shortening, SSC stretch-shortening cycle contraction, VL vastus lateralis, VO_2max maximal oxygen uptake

Two independent reviewers independently assessed the risk of bias in the included studies using the Cochrane Collaboration’s tool [56], containing the following criteria: (1) selection bias; (2) performance bias; (3) detection bias; (4) attrition bias; (5) reporting bias; (6) other sources of bias. Studies were defined as high risk of bias when ≥ 1 of these items was high risk, and low risk if all were low risk. In other situations, it was defined as moderate risk. In this study, the risk of bias in the included literature was visually assessed by Review Manager 5.4.1 software.

Statistical analysis and grading the evidence

Effect size (ES) was calculated as standardized mean difference (SMD; Hedges’s g), with 95% confidence interval (CI). ES was classified as trivial (< 0.2), small (0.2 ~ 0.49), moderate (0.5 ~ 0.79), or large (> 0.8). The meta-analysis was performed in Stata v18.0 (STATA Corp., College Station, TX) using the inverse variance method. Heterogeneity was assessed by measuring the inconsistency (I² statistics) in intervention effects across trials. The level of heterogeneity was interpreted according to the Cochrane Collaboration’s guidelines: trivial (< 25%), low (25 ~ 50%), moderate (50 ~ 75%), and high (> 75%). A random-effects model was used to estimate the pooled effect, accounting for heterogeneity across studies due to differences in participants and intervention characteristics. The publication bias was assessed by the funnel plot and Egger’s test. Subgroup analysis was performed to assess potential sources of heterogeneity. If a significant asymmetry was detected, we used the Trim and Fill method to assess the results’ sensitivity. All statistical significance was set at p < 0.05.

When multiple effect sizes were reported within the same outcome in a study, one was chosen for the primary analysis to reduce dependency. The selection followed a hierarchy: the primary outcome, the most common test, or the outcome most related to the trained exercise. Sensitivity analyses including all effect sizes tested the primary findings’ robustness. Additionally, the quality of evidence for outcomes was evaluated using the Grading of Recommendations Assessment, Development and Evaluation (GRADE), which assesses study limitations, imprecision, inconsistency, indirectness, and publication bias.

Results

The screening flow diagram is shown in Fig. 1. A total of 3157 relevant publications were retrieved (PubMed n = 1874, Web of Science n = 394, Medline n = 395, Sport-Discus n = 490, Manual search n = 4). Of these, 1277 were duplicates, and 1840 were excluded after reviewing the titles and abstracts. After evaluating the full texts, 20 of the 40 publications were removed, leaving 20 studies for analysis.

Fig. 1.

Study flowchart

Participant characteristics

A total of 556 healthy participants with mean ages ranging from 19.0 to 67.6 years were included (Table 2). Sex information was missing in one study. Regarding training status [57], 38 were sedentary and had not participated in regular resistance or strength-based sports. Moreover, 442 participants were recreationally active, which had muscle-strengthening activities 2 or more days a week. 21 were highly trained and achieved national-level performance. 55 were elite, which competed at the international level. Based on resistance training experience, 203 participants were trained and 353 were untrained.

Table 2.

Subgroup analysis results regarding the effects of non-failure training on exercise Performance

Outcomes	Variables	No. of participants	SMD (95%CI)	P value	Test of heterogeneity
					P value	I2(%)
Dynamic Strength	Participant Characteristics
	Trained	178	0.38 (0.09,0.66)	0.01	0.99	0
	Untrained	283	0.09 (-0.07,0.25)	0.19	0.58	11.34
	Training Method
	TRT	385	0.20 (0,0.39)	0.05	0.79	0
	VBT	76	0.47 (0.03,0.91)	0.04	0.96	0
	Training Duration
	≤ 6weeks	83	0.20 (-0.22,0.62)	0.35	0.46	0
	> 6weeks	378	0.25 (0.06,0.45)	0.01	0.86	0
	Training Volume
	Volume equated	256	0.21 (-0.03,0.45)	0.09	0.71	0
	Non-volume equated	205	0.29 (0.02,0.56)	0.03	0.80	0
Isometric Strength	Training Duration
	≤ 6weeks	49	0.05 (-0.49,0.60)	0.85	0.88	0
	> 6weeks	63	-0.26 (-0.74,0.22)	0.29	0.43	0
Muscular Hypertrophy	Participant Characteristics
	Trained	68	-0.14 (-0.69,0.42)	0.62	0.26	29.36
	Untrained	188	-0.16 (-0.44,0.12)	0.27	0.73	0
	Training Method
	TRT	217	-0.04 (-0.30,0.22)	0.75	0.98	0
	VBT	39	-0.79 (-1.41, -0.16)	0.01	0.77	0
	Training Volume
	Volume equated	153	-0.17 (-0.48, 0.14)	0.28	0.57	0
	Non-volume equated	103	-0.12 (-0.50, 0.26)	0.53	0.50	0
Muscular Power	Participant Characteristics
	Trained	137	0.03 (-0.29,0.36)	0.85	0.65	0
	Untrained	32	-0.04 (-0.70,0.62)	0.90	0.77	0
	Training Method
	TRT	131	-0.05 (-0.38,0.29)	0.78	0.75	0
	VBT	38	0.23 (-0.38,0.84)	0.45	0.87	0
	Training Duration
	≤ 6weeks	86	-0.08 (-0.49,0.33)	0.71	0.43	0
	> 6weeks	83	0.12 (-0.30,0.53)	0.59	0.92	0
	Training Volume
	Volume equated	115	-0.02 (-0.37,0.34)	0.93	0.56	0
	Non-volume equated	54	0.09 (-0.43,0.60)	0.74	0.80	0

TRT traditional resistance training, VBT velocity-based training

Training protocol

Non-failure training standards encompass maintaining the same load while decreasing exercise volume, adhering to the same protocol with a reduced load, or allowing voluntary interruption, where subjects opt to cease before reaching muscular failure. Specifically, eighteen studies [12, 14, 22, 24, 25, 41–52, 54] employed the same load but reduced repetitions. One study [55] followed the same protocol but with reduced load, and two studies [41, 53] performed to voluntary interruption. Regarding failure training, eight studies [25, 42, 44, 47–49, 53, 54] were performed to inability to complete the concentric phase of a repetition during each set. Twelve studies [12, 14, 22, 24, 41, 43, 45, 46, 51, 52, 55] implemented the maximum possible repetitions against a given resistance.

Among the included studies, eleven [12, 14, 24, 25, 42, 47–51, 54] used a single exercise, such as knee extension or squat, for resistance training. The remaining ten [22, 41, 43–46, 52, 53, 55, 58] used two or more exercises. Ten studies [14, 24, 25, 41, 42, 47–49, 51, 53] took single-joint exercises, while eleven studies [12, 22, 43–46, 50, 52, 54, 55, 58] adopted multi-joint exercises like bench press and deadlift. Regarding body region (upper or lower limbs), six studies [24, 42, 47–50] used upper body exercise, seven [12, 14, 25, 41, 51, 53, 54] used lower body exercise, and eight studies [22, 43–46, 52, 55, 58] incorporated both methods in the exercise protocol. Twelve studies [12, 14, 22, 25, 43–46, 51, 52, 54, 58] used free-weight exercises, while the remaining used a Smith machine.

Training load largely varied across these studies. The included studies implemented three intensity types: fixed [14, 24, 25, 41, 42, 44, 45, 48, 49, 51, 53–55], incremental [12, 43, 46, 47, 50, 58], and wave-like periodization [22, 52]. For training volume, non-failure groups typically performed 3–40 sets of 1–12 repetitions. The failure group performed 1–4 sets of 2–12 repetitions. Training program durations ranged from 4 to 14 weeks, with a median of 8.7 weeks. Training frequency ranged from 2 to 4 days per week. Moreover, a novel, velocity-based approach has been used to determine exercise termination in three studies [44, 50, 54]. One [50] used 40% velocity loss to represent muscle fatigue.

Outcome measurements

Outcomes included measures of dynamic strength (e.g., isotonic or isokinetic strength), static strength (e.g., isometric strength) muscular hypertrophy, muscular endurance, and muscular power. Muscular strength was assessed in terms of 1RM in eighteen studies [14, 22, 24, 25, 41–45, 47–51, 55, 58–60] and isometric or isokinetic assessments in six studies [14, 24, 25, 42, 47, 51]. The changes in muscle cross-sectional area were used for muscular hypertrophy testing in seven studies [41, 47–51, 53], thickness in three studies [25, 45, 54], volume in one study [50] and circumference in one study [45]. The changes in the MNR were used for muscular endurance testing in four studies [22, 25, 47, 54] Muscular power was assessed in CMJ in five studies [22, 43, 45, 50, 52] and exercise-specific power in four studies [12, 22].

Risk of bias

The results of the quality evaluation of the included 20 studies were shown in Fig. 2. Overall, the included studies showed some methodological concerns. Most studies reported random allocation procedures, although the method of sequence generation was insufficiently described in several trials. Allocation concealment was the most frequently unclear domain, as many studies did not provide sufficient information on whether group assignment was concealed before allocation. Because of the nature of resistance training interventions, blinding of participants and personnel was generally not feasible, potentially introducing performance bias. Blinding of outcome assessors was inconsistently reported across studies, and this may be particularly relevant for outcomes requiring maximal voluntary effort or assessor-dependent measurements, such as strength testing and muscle thickness assessment. Incomplete outcome data and selective reporting were generally judged to pose low or unclear risk when attrition and outcome reporting were adequately described; however, several studies did not provide sufficient information for a definitive judgment. These methodological limitations were considered when interpreting the pooled estimates and when rating the certainty of evidence using the GRADE framework.

Fig. 2.

Risk of bias in the included studies

Meta-analysis

We conducted subgroup analyses of participant characteristics (trained vs. untrained), training volume (equal vs. non-equal), training method (traditional RT vs. VBT), and training duration (≤ 6 weeks vs. > 6 weeks) to identify potential moderators of training effects.

Sensitivity analyses

Sensitivity analyses, including all effect sizes, yielded results broadly consistent with the primary analyses, indicating that the main conclusions were not materially affected by the choice of representative effect sizes (Table S3).

Muscular strength

Dynamic strength

Sixteen publications [22, 25, 41, 43–45, 47–51, 53–55, 59, 60] showed that non-failure training increased dynamic strength compared with the failure training group. Three publications [14, 24, 42] found no difference between the groups in terms of increasing dynamic strength.

The pooled ES of dynamic strength was small and significant (SMD_pooled = 0.24, 95% CI: 0.06 to 0.42, p = 0.010, Fig. 3) with trivial heterogeneity (I² = 0%, p = 0.890). The funnel plot and Egger’s test (p = 0.632) indicated no risk of publication bias (Fig. 8).

Fig. 3.

Forest plot of the effects of non-failure training on dynamic strength

Fig. 8.

Publication bias funnel

Subgroup analysis revealed that participant characteristics contributed significantly to the effects of dynamic strength. Specifically, the ES was small and significant in trained populations (SMD = 0.38, 95% CI: 0.09 to 0.66, p = 0.010), while it was trivial and not significant in untrained populations (SMD = 0.09, 95% CI: -0.08 to 0.41, p = 0.190). Regarding the training volume, the ES was small and significant in non-equal volume (SMD = 0.29, 95% CI: 0.02 to 0.56, p = 0.030), while it was trivial and not significant in equal volume (SMD = 0.21, 95% CI: -0.03 to 0.45, p = 0.090). For the training method, traditional RT yielded a small and significant ES (SMD = 0.20, 95% CI: 0 to 0.39, p = 0.050), VBT also produced a small and significant ES (SMD = 0.47, 95% CI: 0.03 to 0.91, p = 0.040). Training duration also influenced dynamic strength. The ES was small and significant in > 6 weeks (SMD = 0.25, 95% CI: 0.06 to 0.45, p = 0.010), and non-significant in ≤ 6 weeks (SMD = 0.20, 95% CI: -0.22 to 0.62, p = 0.350).

Isometric strength

Three publications [14, 42, 47] indicated that non-failure training increased isometric strength compared to the failure training group, whereas two publications [24, 51] found no difference between the groups in isometric strength.

The pooled ES of isometric strength was trivial and not significant (SMD_pooled = -0.12, 95% CI: -0.48 to 0.24, p = 0.510, Fig. 4) with no heterogeneity (I² = 0%, p = 0.660). The funnel plot and Egger’s test (p = 0.821) indicated no publication bias (Fig. 8).

Fig. 4.

Forest plot of the effects of non-failure training on isometric strength

Subgroup analyses indicated that participant characteristics, training volume, training method, and training duration were not significant moderators of isometric strength.

Muscular hypertrophy

Five [41, 45, 47, 49, 53] showed that training to failure increased muscular hypertrophy compared to non-failure training, whereas Five publications [25, 48, 50, 51, 54] found no significant difference between the groups in muscular hypertrophy.

The pooled ES of muscular hypertrophy was trivial and not significant (ES_pooled = -0.15, 95% CI: -0.39 to 0.09, p = 0.220, Fig. 5) with no heterogeneity (I² = 0%, p = 0.710). The funnel plot and Egger’s test (p = 0.884) indicated no publication bias (Fig. 8).

Fig. 5.

Forest plot of the effects of non-failure training on muscular hypertrophy

Subgroup analyses indicated that participant characteristics, training volume, training method, and training duration were not significant moderators of muscular hypertrophy.

Muscular endurance

Two publications [22, 54] showed that training to failure significantly increased muscular endurance compared to non-failure training, while another two publications [25, 47] found no significant difference between the groups to increase muscular endurance.

The pooled ES of muscular endurance was trivial and not significant (SMD_pooled = -0.02, 95% CI: -0.45 to 0.41, p = 0.930, Fig. 6) with low heterogeneity (I² = 30.96%, p = 0.330). The funnel plot and Egger’s test (p = 0.489) indicated no publication bias (Fig. 8).

Fig. 6.

Forest plot of the effects of non-failure training on muscular endurance

Subgroup analyses indicated that participant characteristics, training volume, training method, and training duration were not significant moderators of muscular endurance.

Muscular power

Five publications [22, 43, 45, 50, 60] showed that non-failure training increased muscular power compared to failure training group; Two publication [12, 54] showed that training to failure increased muscular power.

The pooled ES of muscular power was trivial and not significant (SMD_pooled = 0.02, 95% CI: -0.28 to 0.31, p = 0.910, Fig. 7) with no heterogeneity (I² = 0%, p = 0.860). The funnel plot and Egger’s test (p = 0.803) indicated no publication bias (Fig. 8).

Fig. 7.

Forest plot of the effects of non-failure training on muscular power

Subgroup analyses indicated that participant characteristics, training volume, training method, and training duration were not significant moderators of muscular power.

GRADE assessment

The quality of evidence for each outcome was rated as moderate based on the GRADE criteria. Detailed evaluations using the GRADE framework, including assessments of risk of bias, inconsistency, imprecision, indirectness, and publication bias (Table S1).

Discussion

The systematic review and meta-analysis compared the effects of RT performed to non-failure versus failure on exercise performance in healthy adults. The main findings were that RT performed to non-failure is an effective method for improving dynamic strength in healthy adults, but it did not show a significant advantage for isometric strength, muscular hypertrophy, endurance, or power compared with RT performed to failure. Participant characteristics, training volume, training method, and training duration are important moderating variables that influence the effectiveness of dynamic strength. These findings suggest that no-failure may be a more effective strategy for optimizing dynamic strength.

The meta-analysis demonstrated that RT performed to non-failure significantly enhanced dynamic strength compared with performing to failure. The small advantage of non-failure training for dynamic strength may be partly explained by better fatigue management and preservation of movement quality [61]. Proppe et al. reported that during RT, higher-order motor units are progressively recruited to maintain force output, but excessive fatigue can disrupt motor unit behavior and neuromuscular control [61]. Additionally, non-failure training may attenuate metabolic stress and inflammation, facilitating muscle recovery and subsequent adaptation [62–64] Conversely, failure training could escalate metabolic stress, potentially impairing muscle protein synthesis and satellite cell functionality, thereby constraining strength development [63, 65]. Schoenfeld reported that training to failure induces greater metabolic stress and neuromuscular fatigue, potentially increasing recovery demands between sessions [65]. From a practical perspective, repeated failure training may increase recovery demands and reduce the quality of subsequent training sessions if fatigue is not adequately managed. By contrast, non-failure training may help preserve movement quality and training consistency while still providing an adequate stimulus for dynamic strength adaptations. Thus, rather than suggesting that training to failure is always detrimental, the present findings indicate that routinely reaching failure may not be necessary for optimizing dynamic strength, particularly when fatigue management and technical quality are important considerations [66].The results of the present meta-analysis are partially inconsistent with previous meta-analyses examining resistance training performed to failure. For example, previous systematic reviews generally reported no significant differences between failure and non-failure training for strength development [67, 68]. One possible explanation for this discrepancy is that the present study included a larger number of randomized controlled trials while excluding non-randomized studies. In addition, four of the newly included studies favored non-failure training for strength gains, and these studies typically implemented relatively high training loads (70–95% 1RM), which may partly explain the observed differences. By contrast, isometric strength showed no significant difference. This discrepancy may be due to being outcome specific. Dynamic strength tests are highly dependent on neuromuscular coordination, intermuscular timing, and skill expression during multi-joint movements, whereas isometric strength tests primarily reflect maximal force production under fixed-joint conditions; isometric strength is less affected by such skill-related factors [69, 70] Consequently, in future practice, appropriate intensity, training volume, and interval time should be prescribed to maximize muscle strength gains.

In contrast to dynamic strength, muscular hypertrophy did not differ significantly between failure and non-failure training. This finding is consistent with previous studies that RT performed to failure did not provide additional benefits for muscle hypertrophy [30, 68]. For example, Schoenfeld et al. reported that training to failure did not appear to be necessary for maximizing muscle hypertrophy [71] Non-failure training may still provide a sufficient hypertrophic stimulus by recruiting a large proportion of motor units while minimizing excessive fatigue and recovery demands [71]. In practical terms, the current evidence does not justify the routine use of failure training as a necessary strategy for maximizing hypertrophy in healthy adults.

Similarly, no significant differences were observed between failure and non-failure training for muscular endurance or power. For muscular endurance, this finding may indicate that both training approaches can enhance local fatigue resistance. Improvements in muscular endurance are generally associated with neuromuscular adaptations, increased oxidative capacity, and enhanced metabolic efficiency within the working muscle [71, 72]. Previous studies have suggested that RT can improve local muscular endurance through neural adaptations and metabolic adaptations regardless of whether sets are performed to failure [71, 72]. Therefore, the presence or absence of momentary muscular failure may be less critical than the overall structure of the training stimulus, including training volume, load, and frequency. From a physiological perspective, the absence of a significant difference between conditions is also plausible for muscular power. Power-oriented adaptations rely heavily on the ability to produce force rapidly, which is strongly influenced by contraction velocity, neural drive, and the rate of force development [34, 73]. Previous studies have shown that excessive fatigue during resistance training can reduce movement velocity and impair neuromuscular performance, potentially limiting power-related adaptations [34, 50]. Consequently, training strategies that avoid excessive fatigue may be beneficial for maintaining movement velocity and optimizing power development. However, the present analysis included a relatively small number of studies examining muscular power, and the outcome measures varied substantially across studies, including both jump-based assessments and exercise-specific power tests. These methodological differences may have reduced the meta-analysis’s sensitivity to detecting small between-group effects. Therefore, current evidence remains insufficient to determine whether one approach provides a clear advantage for power development.

The subgroup findings for dynamic strength offer additional, yet still cautious, insights. The pooled ES appeared more favorable in trained participants than in untrained participants. A plausible explanation is that trained individuals may require a more precise balance between stimulus and fatigue to continue progressing [59, 73]. Because they already possess a higher degree of neuromuscular adaptation, unnecessarily accumulating fatigue through repeated failure may impair performance quality without providing additional adaptive benefit [50, 74]. In contrast, untrained individuals often respond positively to a wide range of resistance training stimuli, which may reduce the detectable difference between failure and non-failure approaches during the early stages of training [59, 75]. Likewise, the pattern of results suggested that non-failure training may be especially useful when programs extend beyond six weeks and when traditional resistance training methods are used. However, these subgroup findings should be interpreted conservatively, as subgroup analyses are observational in nature and may be underpowered. They are better viewed as hypothesis-generating than definitive evidence of moderation.

From a practical perspective, non-failure training may be considered when the goal is to improve dynamic strength while managing fatigue and maintaining movement quality. This implication may be particularly relevant for trained individuals or athletes who need to balance resistance training with sport-specific practice, recovery demands, and readiness for subsequent sessions. However, these recommendations should be interpreted as practical implications rather than direct empirical conclusions, because most included studies did not directly assess training periodization, athlete readiness, or long-term fatigue management. For untrained individuals, both failure and non-failure training may provide sufficient stimulus during the early stages of adaptation. For highly trained or elite athletes, the applicability of these findings should be considered in relation to sport-specific demands, training phase, recovery status, and the overall program design. Failure training may still be useful in specific contexts, but the current evidence does not support its routine use as a superior default strategy across all neuromuscular outcomes.

Limitations

Several limitations should be acknowledged. First, most of the included studies had small sample sizes, which may have limited their statistical power and robustness of the estimated effect sizes. Second, there was considerable variability in how the training protocols were implemented across the included studies, including differences in training frequency, duration, and load intensity. These inconsistencies may have affected the pooled outcomes. Additionally, the participant characteristics varied widely, ranging from recreationally active individuals to competitive athletes across various sports. However, few studies conducted subgroup analyses to explore potential differences in responses among these diverse participants. Meta-regression wasn’t conducted due to few studies and low heterogeneity, limiting the ability to identify independent predictors of variability.

Conclusions

The meta-analysis suggests that resistance training performed to non-failure demonstrates superior efficacy in improving dynamic strength compared with training to failure, while no significant differences were observed for isometric strength, muscular hypertrophy, muscular endurance, or muscular power. These findings indicate both the potential and the limitations of non-failure training in exercise performance enhancement. Future research should explore the long-term physiological and neuromuscular adaptations associated with non-failure training across different populations, training statuses, and sport events. Such work will be essential to determine the practical applicability and efficacy of non-failure as a strategic component of contemporary strength and conditioning programs.

Abbreviations

RT

Resistance Training

1RM

One-Repetition Maximum

VBT

Velocity-Based Training

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

PROSPERO

International Prospective Register of Systematic Reviews

MVC

Maximal Voluntary Contraction

CSA

Cross-Sectional Area

ES

Effect Size

SMD

Standardized Mean Difference

CI

Confidence Interval

I^²

Inconsistency Index

GRADE

Grading of Recommendations Assessment, Development and Evaluation

MNR

Maximum Number of Repetitions

CMJ

Countermovement Jump

Authors’ contributions

S.W., L.X., and Z.Z. had full access to all the data in the study and took responsibility for the integrity of the data and the accuracy of the data analysis; Concept and design: S.W., L.X, and Z.Z.; Acquisition, analysis, or interpretation of data: X.Z., X.Y., and J.H.; Drafting of the manuscript: K.Z. and P.T.; Critical revision of the manuscript for important intellectual content: All authors; Statistical analysis: K.Z., X.Y., and P.T.

Funding

This research was supported by the Chongqing Talents Program (Grant No. 524Z240323).

Data availability

The datasets analyzed during the current study are available in the Science Data Bank (Science DB) repository at https://doi.org/10.57760/sciencedb.31628.

Declarations

Ethics approval and consent to participate

Not applicable.

Competing interests

The authors declare no competing interests.