How Accurate Are 1RM Calculators? (And How to Use Them Anyway)

TL;DR: Rep-max formulas are estimates dressed up as precision. They're reasonably tight when you feed them a heavy set of 5 or fewer, and they fall apart on high-rep sets because people vary enormously in how many reps they can grind out at a given percentage. Pick one formula, keep your test sets short and honest, and treat the output as a yardstick for change rather than a number to put on a competition platform.

You punch 225 for 8 into a calculator and it hands back 279. Confident. Three significant figures. It feels like a measurement.

It isn't. It's a curve fit built on other people's bodies, and it has a known error bar that most apps never show you. That doesn't make it useless. It makes it a tool with a manual, and almost nobody reads the manual.

Where the number comes from

Every rep-max calculator is running some version of the same idea: there's a predictable relationship between the weight you lift, the reps you get, and the weight you could lift for exactly one. Epley's formula (1985) and Brzycki's (1993) are the two you've almost certainly used, and they're both simple algebra fitted to a modest set of observations.

The classic accuracy test is LeSuer et al. (1997), who ran seven of these equations against real 1RM tests in the bench press, squat, and deadlift. Correlations were excellent, north of 0.95 in most cases, which is the statistic everyone quotes. The more useful finding is the one people skip: the equations didn't perform equally across lifts, and every single one of them underestimated the deadlift.

That's the first thing to internalize. A high correlation means the formula ranks lifters correctly. It does not mean the formula gives you the right number.

Why high-rep sets break the math

The deeper problem is the assumption underneath all of these formulas: that a given percentage of your 1RM corresponds to a roughly fixed number of reps. Seventy percent means twelve. Everyone knows the chart.

The chart is wrong, and we now know by how much. Nuzzo and colleagues (2024) pooled 952 reps-to-failure tests across 7,289 people in a meta-regression published in Sports Medicine. Their point estimates land near 14 reps at 70 percent, 9 at 80 percent, and 4 at 90 percent for the bench press. For the leg press, the same percentages produced roughly 19, 13, and 9 reps.

Read those two lines again. Same percentages, wildly different rep counts, purely because the exercise changed. Exercise was the only moderator that meaningfully shifted the relationship in their analysis. Sex, age, and training status barely mattered.

And the spread between individuals is worse than the averages suggest. Richens and Cleather (2014) put eight competitive weightlifters and eight endurance runners on a leg press at 70 percent of 1RM. The weightlifters averaged about 18 reps. The runners averaged about 40. Same relative load, more than double the reps, because muscle fiber makeup and fatigue resistance are not standard equipment.

So when a calculator converts your set of 15 into a one-rep max, it's quietly assuming you're the average of a population you may sit nowhere near. Feed it a set of 3 and that assumption barely has room to hurt you. Feed it a set of 15 and the assumption is doing most of the work.

The rule that fixes most of the error

Keep your estimation sets at five reps or fewer.

That's it. That's the intervention with the highest return. The shorter the extrapolation, the less the formula has to invent. A set of 3 is a small step from a single. A set of 12 is a leap across the exact territory where humans differ most from each other.

Two more things tighten it up:

Take the set genuinely close to failure. An estimate built on a set you stopped four reps early is an estimate of a lift you didn't do. If you're not sure how close to the edge you actually were, that's a calibration problem worth fixing directly, and we've covered how to calibrate reps in reserve properly.

Use the same formula every time. Epley and Brzycki disagree with each other, and the gap widens as reps climb. Which one is "right" matters far less than never switching, because switching mid-training-block manufactures progress or erases it out of thin air.

Compared to what, exactly

There's a fair objection here: if the estimate is soft, why not just test a true single?

Because true 1RM testing isn't as clean as it sounds either. Grgic et al. (2020) reviewed the test-retest reliability of 1RM assessment and found trained lifters reproducing their maxes with coefficients of variation from 0.5 to 7.8 percent. At the top of that range, a 400 pound deadlift is a 400 pound deadlift on Tuesday and a 369 pound deadlift on Friday, with no strength lost in between. Sleep, food, caffeine, warmup quality, and how much you wanted it all move that number.

A max attempt is also expensive. It costs a session, it costs recovery, and it carries injury risk that a heavy triple doesn't. You can't run one every week, which means you can't use it to steer training.

So the real comparison isn't "estimate versus truth." It's a slightly noisy number you can generate every session against a slightly noisy number you can generate four times a year. The estimate wins on utility by a mile.

Use it as a yardstick, not a verdict

Here's the reframe that makes all of this workable: the absolute value doesn't matter much, and the change does.

If your estimated bench 1RM goes from 250 to 268 over eight weeks, the systematic error in the formula is sitting in both numbers. It largely cancels. You didn't learn your true max, but you learned something better, which is that you got stronger and roughly by how much. That's the signal you actually need to decide whether the block worked, whether to push volume, or whether it's time to take a deload.

This is also why estimated maxes are the honest engine behind progressive overload. Adding five pounds to the bar means nothing if your reps quietly dropped from 8 to 5. The estimate collapses load and reps into one number, so you can see whether the work is compounding or just rearranging itself.

It's the same logic behind the training max in 5/3/1. Wendler doesn't ask for your true max. He asks for 90 percent of it, precisely because the true number is unstable and building a program on an unstable input breaks the program.

What to actually do

Test with a heavy set of 3 to 5, taken to within a rep or two of failure, on the same lift, with the same setup, every three or four weeks. Run it through one formula and never change formulas. Write it down. Watch the line.

If you want the arithmetic done for you, our one-rep max calculator averages Epley and Brzycki, which smooths out the disagreement between them, and hands back a percentage chart you can program off directly.

And expect the deadlift number to read low. Every equation LeSuer's group tested underestimated it, and yours probably will too.

This is roughly how Protokl handles it under the hood. Every working set you log gets converted to an estimated 1RM using one formula and never a mix, sets above ten reps are excluded from the capability calculation because that's where the estimate stops meaning much, and the number the app trusts is your best recent estimate rather than whatever happened in your last session. Your prescribed top set is anchored to that, so the bar follows the estimate up when you're progressing and the program responds when you stall, without you running any of this arithmetic yourself.

References

LeSuer DA, McCormick JH, Mayhew JL, Wasserstein RL, Arnold MD (1997). The accuracy of prediction equations for estimating 1-RM performance in the bench press, squat, and deadlift. Journal of Strength and Conditioning Research, 11(4), 211-213.
Nuzzo JL, Pinto MD, Nosaka K, Steele J (2024). Maximal number of repetitions at percentages of the one repetition maximum: a meta-regression and moderator analysis of sex, age, training status, and exercise. Sports Medicine, 54, 303-321.
Richens B, Cleather DJ (2014). The relationship between the number of repetitions performed at given intensities is different in endurance and strength trained athletes. Biology of Sport, 31(2), 157-161.
Grgic J, Lazinica B, Schoenfeld BJ, Pedisic Z (2020). Test-retest reliability of the one-repetition maximum (1RM) strength assessment: a systematic review. Sports Medicine - Open, 6, 31.