How accurate is a one rep max estimate?
Accurate enough at low reps, unreliable at high ones. From a set of three, four standard formulas land within about six kilos of each other on a 100 kg lift. From a set of twelve, they spread by nearly sixteen.
Updated 2026-08-24
What these formulas actually are
A one rep max is the heaviest weight you can lift once. Estimating it from a lighter set means asking: if this person managed a hundred kilos for five reps, what could they manage for one?
The common answers are curve fits, not physics. Epley multiplies the weight by one plus the reps divided by thirty. Brzycki divides by thirty-seven minus the reps. Lombardi raises the rep count to the power of 0.1. O'Conner uses one plus reps divided by forty. Each was fitted to a set of observed lifters, and each describes that population reasonably well.
None of them models anything about muscle. They are descriptions of an average relationship between load and repetitions, which is why they diverge as you move away from the data they were fitted to — and why no amount of argument about which is best will produce a right answer for an individual.
The practical consequence is that the choice of formula matters much less than the input you feed it. A careful set of three through any of them beats a sloppy set of ten through the theoretically best one.
Where they agree and where they fall apart
Run all four on a hundred kilos and the pattern is clear. At three reps they return 110.0, 105.9, 111.6 and 107.5 — a spread of 5.7 kilos, or under six percent. At five reps the spread is 5.0. At eight it is 6.7.
At ten reps the spread widens to 8.3. At twelve it reaches 15.8 — Brzycki says 144.0 while Lombardi says 128.2. Those are not small disagreements about a detail; they are different training programmes.
The reason is structural. All four were fitted mainly to low-rep data, where the relationship between load and reps is close to linear and well observed. Extrapolating them into high-rep territory pushes each formula's particular shape into a region it was never calibrated for, and the shapes differ.
Brzycki shows this most starkly because of its form: dividing by thirty-seven minus the reps means the denominator shrinks as reps rise, so the estimate climbs steeply and would divide by zero at thirty-seven reps. It is accurate at low reps and increasingly aggressive as the rep count grows.
The number that actually matters is the rep count
Everything above points to one practical rule: test with a low rep count. Somewhere between three and six is the useful range, and the estimate degrades steadily above it.
This is convenient rather than restrictive, because a heavy set of three is also far safer and less disruptive than an actual one rep max attempt. A true maximal single carries genuine injury risk, requires a spotter on most lifts, and produces enough fatigue to compromise the following days of training. A triple gives you almost the same information at a fraction of the cost.
The reps also have to be honest. These formulas assume a set taken close to failure with clean technique. A set of eight that could have been eleven produces an underestimate; a set of eight where the last three were form breakdown produces an overestimate that will make every derived working weight too heavy.
Above about twelve, treat the output as a rough guide rather than a number. Endurance and pain tolerance start to determine high-rep performance as much as maximal strength does, and those are not what the formulas were fitted to measure.
What the estimate is for
Most training programmes prescribe work as a percentage of one rep max, which is what makes the estimate useful even though it is imprecise.
The standard percentages run roughly: 95 percent for two reps, 93 for three, 90 for four, 87 for five, 85 for six, 80 for eight, 75 for ten and 70 for twelve. Those are population averages too, and individuals vary — some people can grind out far more reps at a given percentage than others, which is a real and fairly stable individual trait rather than a training deficiency.
Because everything downstream is a percentage of the estimate, an error in the estimate propagates proportionally. A one rep max overestimated by ten percent makes every working set ten percent too heavy, which turns a programme of controlled progression into a programme of missed reps.
This is why the safer direction for the error matters. If you are going to be wrong, be wrong low — an underestimate produces sets that feel easy and progress that continues, whereas an overestimate produces failed sessions and the fatigue that comes with them. Taking the more conservative of several formulas is a defensible default rather than false modesty.
When to re-test, and when not to bother
An estimate ages. Strength changes, and a percentage-based programme running off a three-month-old number is prescribing the wrong weights in one direction or the other.
Re-testing every four to six weeks, or at the end of a training block, is a reasonable rhythm. Do it with the same lift, similar conditions and the same rep count, because comparing a triple in one block against a set of eight in the next introduces exactly the formula divergence described above and can show progress or regression that did not happen.
There is also a case for not using a calculated max at all. Autoregulated approaches — rating how hard a set felt, or how many reps you had left in reserve — adjust to how you are performing today rather than to a number derived weeks ago. Sleep, stress, nutrition and accumulated fatigue move actual daily capacity considerably, and no formula sees any of that.
The reasonable synthesis is to use the estimate to set the shape of a programme and daily judgement to set the load within it. The calculation tells you roughly where you are; only the bar tells you what today is.