Blog / Training

Training to Failure: When It Helps and When It Costs You

November 8, 2026 · 11 min read

One camp says a set only counts if the last rep fails; the other says failure wrecks recovery and should be avoided. Research from the last few years answers most of it, and the answer depends on what you are training for and which exercise is in your hands.

What "failure" means

Researchers and lifters use the word for different things, and a lot of the argument comes from mixing them up.

• Momentary muscular failure: you try the next rep with full effort and the weight does not move through the full range. This is what most studies mean when they say "failure".
• Technical failure: the point where you can still grind out a rep, but only by changing how you move. Hips shoot up on a squat, the bar drifts forward on a bench, the elbows flare on a row. Many coaches treat this as the real stopping point on big lifts.
• Stopping short: ending the set with reps you could still do, usually counted as reps in reserve (RIR). A set at 2 RIR means you think you could have done two more good reps. The RIR guide covers the scale in detail.

Proximity to failure is the umbrella term for all of this: how many reps from failure a set ends. It helps to think of proximity as a scale that runs down to 0.

What the research says about muscle growth

Several meta-analyses have compared failure with non-failure training; two recent ones stand out. Grgic and colleagues (2022) pooled 15 studies in young adults and found no significant difference in muscle growth between the two approaches overall. In the subgroup of resistance-trained participants, failure training showed a small advantage. Refalo and colleagues (2023) pooled 15 hypertrophy studies with a stricter look at how each study defined failure. Pooling every definition of failure, they found a trivial edge for failure; restricted to momentary muscular failure, there was no advantage either way. They also noted that pushing sets harder did not always produce more growth, which hints at a curve that may flatten near the top.

A within-subject trial gives a cleaner picture. Refalo and colleagues (2024) had 18 trained men and women train the leg press and leg extension twice a week for 8 weeks, one leg to momentary failure and the other leg stopping at about 1-2 RIR. Quadriceps thickness increased by almost exactly the same amount in both legs. The failure leg lost more reps and more bar speed across its sets, so it carried more fatigue for the same result.

The research that adds nuance comes from Robinson and colleagues (2024), who estimated the RIR of every training condition across a large body of studies and ran a series of meta-regressions on it. Muscle growth improved as sets were taken closer to failure. The authors called the analysis exploratory, because the RIR values were estimated from study descriptions and the model fit was modest, but the direction fits the rest: getting close matters, and the last rep or two before failure may add a little.

Put together, sets that end a few reps short of failure grow muscle about as well as sets taken all the way. Sets that end far from failure, say 5 or more reps in reserve, likely grow less.

What the research says about strength

Strength responds differently. In the Robinson meta-regressions, strength improved about equally across a wide range of RIR; ending sets nearer to failure made little difference. In the Grgic review, when the two groups were not matched for volume, non-failure training produced better strength results.

Velocity-based training studies show the same pattern from another angle. Pareja-Blanco and colleagues (2017) had two groups squat for 8 weeks, one ending each set once bar speed dropped 20% and the other at a 40% drop, which is much closer to failure. Squat strength improved similarly in both groups, and the 20% group improved its jump more while doing 40% fewer reps. The 40% group grew more in two of the quadriceps muscles and shifted away from its fastest (type IIX) fiber type, while the 20% group kept it.

Strength is partly a skill: moving heavy weight with consistent technique. Sets that end with fast, clean reps practice that skill. Grinding reps, where the bar slows and form drifts, practice a worse version of the lift.

The cost of a failure set

Failure sets get the same result at a higher cost.

Morán-Navarro and colleagues (2017) had trained men do bench press and squat in three ways: 3 sets of 5 with a load they could lift 10 times, 6 sets of 5 with the same load, and 3 sets of 10 to failure. The 6 x 5 and 3 x 10 protocols had the same total reps. The failure protocol produced a bigger immediate drop in jump height and bar speed, and its performance and fatigue markers were still lagging 24-48 hours later, a window in which both non-failure protocols had recovered faster.

Failure also eats into the sets that follow it in the same session. Here is an illustration with 3 sets of leg extensions at 100 lbs, two minutes apart, by someone whose best set at that weight is about 15 reps:

• Every set to failure: 15, then 11, then 9. That is 35 reps, and the last set is grim.
• Every set stopped at 1 RIR: 14, then 13, then 12. That is 39 reps, and the lifter walks away able to train legs again a couple of days later.

The numbers will differ for you, but the shape is common: the first failure set buys one extra rep and costs several reps in the sets after it. Over a week, those lost reps and the slower recovery can mean less total work, and weekly volume is one of the main drivers of muscle growth. In the 8-week leg trial, total volume ended up similar in both legs, so the clearer cost there was fatigue.

Failure sets also feel worse. In a follow-up paper on the same 8-week trial (Refalo and colleagues, 2025), participants rated the failure sets as more uncomfortable and the sessions as harder, and reported more positive feelings after stopping short. A program you dread is harder to keep running for months.

Where failure is worth using

Failure sets are worth it in three places.

The last set of safe exercises

On machines, cables, dumbbell curls, lateral raises, leg extensions and similar lifts, a failed rep ends with the weight lowered under control. Technique barely changes as you tire. Taking the final set of these to failure costs little, adds a small amount of stimulus, and puts no other set at risk because nothing comes after it.

Calibrating your effort ratings

Lifters tend to underestimate how many reps they have left, by about one rep on average, and they judge less accurately far from failure and on sets longer than about 12 reps (Halperin and colleagues, 2022). An occasional all-out set shows you where your real limit is, which can make your RIR calls more accurate. One set per exercise every few weeks is plenty; the RIR tracking guide walks through a calibration routine.

Bodyweight and light-load work

With light loads or bodyweight exercises like push-ups, it is hard to judge 2 RIR when you can do 30 reps. With light loads, effort seems to matter more: in an 8-week knee extension study in untrained men, sets at 30% of 1RM grew the quadriceps meaningfully only when taken to failure, while sets at 80% grew it either way (Lasevicius and colleagues, 2022). Going close to or all the way to failure makes more sense here than on a heavy set of 5.

Where failure costs more than it gives

Heavy barbell compounds are the clearest case. On squats, deadlifts and bench press, form often breaks before the muscle gives out, so a set "to failure" turns into a set of technique failure under a heavy load. A failed squat or bench without safeties or a spotter is a real injury risk. These lifts also take the most whole-body recovery, so the slower recovery shows up across the rest of your week.

Strength-focused blocks are the second. Since strength does not seem to need failure, ending sets 2-4 reps short keeps the bar moving fast and the technique sharp, and leaves room to add weight next week.

Beginners are the third. In your first months the main job is learning the movements, and almost any reasonable effort produces progress. A beginner benefits more from clean reps at 2-3 RIR plus the occasional calibration set than from grinding every set.

A practical default

If you want one set of rules to start from:

• Heavy compound lifts (roughly 1-6 reps): stop 2-4 reps short. Never fail them without safeties.
• Moderate-rep compound work (roughly 6-12 reps): stop 1-3 reps short.
• Isolation and machine work: stop 0-2 reps short, with the last set allowed to reach failure.
• Testing: one all-out set per exercise every few weeks to check your RIR calls.

Adjust by how you recover. If sessions keep feeling heavier at the same weights, your sleep is slipping and your logged effort is creeping up, you are probably pushing too many sets too close to the line, and the signs of a deload are worth checking. If every set ends with an easy 4-5 reps left, you are probably stopping too early for muscle growth.

The calculator below turns a target effort into a weight. Enter a recent hard set, then the reps and RIR you want for the next working set, and it estimates a load to start with. Adjust on the day if the first set lands harder or easier than planned.

All of this runs on RIR. If "stop 2 reps short" feels vague, the guide to autoregulation with RPE and RIR shows how to use it to set the day's weight.

How Anneal handles effort

Anneal puts an RIR field on every working set by default, so how close you went to failure is logged next to the weight and reps. Two of the five fatigue signals it reads are about effort: sets that keep grinding near failure, and RIR that stays low across your sessions. When one of them crosses its threshold, Anneal suggests a recovery week and names the signal that fired. An optional setting, off by default, goes further inside a session: when you log a set very close to failure, it makes the next set of that exercise a step lighter, and you can change it back.

Common questions

Do you have to train to failure to build muscle?

No. Two meta-analyses that compared failure with non-failure training (Grgic and colleagues, 2022; Refalo and colleagues, 2023) found little or no difference in muscle growth, and an 8-week trial in trained lifters found the same quadriceps growth from sets taken to failure and sets stopped 1-2 reps short (Refalo and colleagues, 2024). What does matter is getting reasonably close: a 2024 analysis across many studies found muscle growth improving as sets ended nearer to failure, so most hypertrophy sets belong 1-3 reps short of failure, with the last isolation set allowed to reach it.

Is training to failure bad for strength?

You can build strength without it, and it slows recovery. Strength improved about equally across a wide range of reps in reserve in the 2024 meta-regressions by Robinson and colleagues, and when volume was not matched, non-failure training came out ahead for strength in the Grgic review. In one trial, sets to failure slowed recovery for up to 24-48 hours. For strength work, stopping 2-4 reps short keeps bar speed and technique while still building strength.

Which exercises are safe to take to failure?

Ones where a failed rep ends harmlessly and technique stays stable as you tire: machines, cables, most dumbbell isolation work, and bodyweight moves like push-ups or chin-ups. Heavy barbell squats, deadlifts and bench presses are poor candidates, because form tends to break before the muscle gives out and a failed rep needs safeties or a spotter. If you do want to test failure on a barbell lift, set the safety pins first.

How many sets should go to failure?

A few, and mostly the last set of an exercise. Taking the final set of an isolation exercise to failure, or one AMRAP set every couple of weeks to check your effort ratings, gives you the information and the stimulus at a small fatigue cost. Taking every set of every exercise to failure drives the number of reps you can do in later sets down and stretches recovery, which usually means less total work across the week.