Best 1RM formula: which to trust for bench, squat, and deadlift

The four common 1RM prediction formulas — Epley, Brzycki, Lombardi, and O'Conner — agree on a 100 kg × 5 reps set within about 5 kg. They disagree by 8 kg or more once you push past 10 reps. LeSuer 1997 tested all four (plus three more) against measured 1RMs in the bench press, squat, and deadlift, and the per-lift answer is Epley for bench, any of the four for squat, and any of the four plus ~10% for deadlift.

Most online 1RM calculators surface a four-formula average. That hides what LeSuer measured: each formula has a different per-lift bias, and averaging shrugs them all together. The bias on bench is small (1–6% under). The bias on deadlift is real (9–14% under across all four). This page walks through what the validation studies actually said, gives a per-lift verdict, and pins the worked-example numbers to the live 1RM calculator so you can verify them against the engine.

The four formulas, side-by-side

Each formula is a curve fit to submaximal-rep data, published between 1985 and 1993. None is empirical in the sense of being derived from a biomechanical model — they are pragmatic algebraic shapes that happened to match the data their authors had in front of them. The per-formula behavior is what differs:

Run a 100 kg × 5 reps set through each formula and the spread is small:

5 kg of spread on a 5-rep estimate. Most online 1RM calculators just spit out the four-formula average and stop there. That's lazy when LeSuer 1997 already told us which formula works for which lift — averaging hides the per-lift differences instead of using them.

What LeSuer 1997 actually found

The validation study most strength references point to is LeSuer, McCormick, Mayhew, Wasserstein, and Arnold, The Accuracy of Prediction Equations for Estimating 1-RM Performance in the Bench Press, Squat, and Deadlift (J Strength Cond Res11(4):211–213, 1997). 67 untrained college students (40 men, 27 women) completed real measured 1RMs and a separate reps-to-fatigue session at the predicted submax weight. Seven prediction equations got compared against the measured 1RM. The headline result holds across all of them: every formula correlated tightly with the achieved 1RM (r > 0.95 for all seven on every lift). The disagreement is in calibration, not in tracking.

Here's the bench-press row from LeSuer's Table 3, restricted to the four formulas this page compares (the original table tested three others — Lander, Mayhew, Wathan — that most consumer calculators don't surface):

Bench-press achieved 1RM mean: 137.3 lb (SD 62.1). LeSuer Table 3, rows 1–4 + 6 (Lander/Mayhew/Wathan rows omitted). All four differences from achieved were significantly different from zero except Mayhew and Wathan (both not in this table).

The deadlift is where the formulas all break the same way:

Deadlift achieved 1RM mean: 236.5 lb (SD 91.7). Every formula in LeSuer's table (all seven, not just the four shown) significantly underestimated the deadlift 1RM.

If you're estimating a deadlift 1RM from a 5RM, add 10% to whatever the calculator returns and call it a day. The math isn't lying about deadlift; the formulas just don't catch up to what a trained lifter holds at 5 reps in a pull from the floor.

Bench press: Epley by ~1%. Brzycki for very low reps.

Epley's linear shape — w × (1 + reps/30), equivalent to adding 3.33% per rep — happens to fit submaximal bench-press data better than any of the other three common forms in LeSuer's sample. 1% under achieved on a 67-subject sample with r = 0.993 is as tight as a four-decade-old napkin formula has any right to be.

The case for Brzycki is at the very-low-rep end. Brzycki's curve rises sharply between 1 and 3 reps and flattens past 10, which fits the “largest drop in % 1RM occurs between 1 and 2 repetitions” finding LeSuer reports in the discussion. LeSuer didn't test formulas in a 2RM-only condition, so the case for Brzycki at 2–3 reps is a structural argument, not a measured one. For the 4–8 rep range most lifters actually program in, default Epley.

Brzycki has a sharper curve at 2–3 reps that LeSuer's data doesn't resolve cleanly — they tested at 10-and-under reps-to-fatigue, but at the low end of that range, not specifically at 2RM. For 4–8 reps on bench, Epley. For a 2RM-derived estimate, Brzycki. At reps = 1, you're not estimating — that IS your 1RM, and the calculator short-circuits all four formulas to return the input weight unchanged.

Squat + deadlift: trust the reliability band, then add for deadlift

On the squat, all four common formulas underpredicted achieved 1RM by 3–5% in LeSuer's sample, and all four had r > 0.96. The differences between them on squat are roughly the same size as the measurement noise on a single lifter from one session to the next. Pick whichever formula your calculator defaults to and read the reliability band — that's the actionable signal. The StrengthMath methodology page documents the band logic in full: ±2% at ≤5 reps, ±5% at 6–10 reps, ±10% at 11–15, ±15% past that.

Deadlift is the lift where the formulas' bias actually changes your decision. All four under-predict by 9–14%. The mechanical explanation in the strength-coaching literature is that a pull from the floor lets a trained lifter hold reps at a higher %1RM than press-from-stop lifts do — there's less elastic loading/unloading per rep, and the bar is fully un-loaded between reps if the lifter sets up clean each pull. Whatever the cause, the adjustment is the simplest possible: take the calculator's deadlift output and multiply by ~1.10 as a starting estimate. Then validate at the bar. The 1.10 is StrengthMath methodology — a clean midpoint of the 9–14% range LeSuer documented across all four formulas — not a published correction factor.

Of the seven formulas LeSuer actually tested, only Wathan was not significantly off on squat (and only Mayhew and Wathan on bench). The Wathan equation is more complex and isn't one of the four most consumer calculators surface. The four common formulas are interchangeable on squat, and the reliability band the engine returns matters more than the formula choice.

Lombardi is the worst of the four. Skip it.

Reynolds, Gordon, and Robergs cross-validated the same family of equations on a different population (n = 70, plus a 20-subject cross-validation group; 18–69 years old, mixed training status; flat bench press and Cybex plate-loaded leg press) at the University of New Mexico (J Strength Cond Res20(3):584–592, 2006). Their conclusion on Lombardi is direct: “The nonlinear equations of Lombardi (19) and Mayhew (21) were less accurate than all linear equations.” That holds across both 5RM and 10RM input conditions, on both bench press and leg press.

Lombardi gets cited in textbooks because it has a power function that looks more sophisticated than a linear one. The 2006 cross-validation found it underperforms every linear competitor — including Brzycki, Epley, and O'Conner — on every action they tested. The power-function shape flattens too fast past 5 reps, which biases the estimate low at exactly the rep range submax-rep formulas are most useful for.

If your calculator gives you a Lombardi-driven number, ignore it and re-run with Epley (bench), or with the four-formula average minus Lombardi (squat / deadlift). The StrengthMath 1RM calculator surfaces all four per-formula values so you can see where Lombardi sits in the spread; it's usually the outlier conservative read. Skip it.

What none of these formulas know

The reps the lifter does matter more than the formula chosen. LeSuer's discussion notes that “<10 reps to fatigue were better for estimating 1-RM,” with prediction error growing sharply past that. The reliability band the engine surfaces (±2% at ≤5 reps, ±5% at 6–10, ±10% past) is doing more work than the formula choice.

What no formula can see: pain or injury (lifting through pain is not a calculator question — stop and consult appropriate care), technique drift on a grindy 5-rep set, range-of-motion differences, equipment (Smith machine vs free bar, sleeves vs no sleeves), and the fatigue state the rep set was performed in. A 5RM tested fresh and a 5RM tested after five working sets will give the formula the same numerical input and very different real-world meaning. The full list of un-modeled factors lives on the methodology page, and a similar callout sits next to each calculator.

For dumbbell pressing the math gets one more wrinkle — per-dumbbell-1RM is the primary output, and the conversion to a barbell equivalent is a wide range, not a single number. The dumbbell-bench 1RM calculator surfaces both. For training percentages off your 1RM, the percentage-of-1RM calculator carries the NSCA + ACSM 2026 dual framing.

If you only memorize one rule from this page: trust low reps over a fancy formula. A 3-rep set run through any of the four formulas is closer to your real 1RM than a 12-rep set averaged across all four. Once your reps creep past 10 you're in “programming hint” territory, not in 1RM-estimation territory, and the reliability band on the calculator output is telling you so.

Common questions

Which 1RM formula is most accurate for bench press?

Epley, by about 1%, in LeSuer 1997's validation against measured bench-press 1RMs (n=67). The full bench-press table: Epley predicted within 1% (135.1 lb predicted vs 137.3 lb achieved), Brzycki under by 4%, Lombardi under by 3%, O'Conner under by 6%. All four correlations against achieved 1RM exceeded r=0.99, so the disagreement is in calibration, not in tracking. For 4–8 reps on bench, default Epley.

Why do all four formulas underestimate deadlift?

Pull-from-floor mechanics let trained lifters hold reps at a higher %1RM than press-from-stop lifts do. LeSuer 1997 found all four common formulas underestimated deadlift by 9–14% across the same sample where bench-press prediction was within 6% (Brzycki 12% under, Epley 10%, Lombardi 11%, O'Conner 14%). The practical rule: take the calculator's deadlift output and multiply by ~1.10 as a starting estimate, then validate at the bar.

Should I average the four formulas or pick one per lift?

Per lift. Bench: Epley. Squat: any of the four — the reliability band matters more than which formula you pick. Deadlift: any of the four × 1.10. Averaging hides the per-lift differences LeSuer 1997 documented across n=67 lifters and that subsequent validation studies (Reynolds-Gordon-Robergs 2006, n=70) confirmed. Most calculators average because it's safe; per-lift picks are more accurate.

Are 1RM formulas accurate enough to use for training programming?

Yes, within the reliability band. The StrengthMath engine returns ±2% at ≤5 reps, ±5% at 6–10 reps, ±10% at 11–15 reps, ±15% past that — bands grounded in LeSuer's directional findings. For a peaking attempt or a meet, run an actual 1RM. For programming a 5×5 or a 4×6 from a 5RM input, the formula is more reliable than a fatigue-dependent 1RM test.

Where to next

Once you have a 1RM number you trust, the next decision is what percentage of it to train at — strength, power, hypertrophy, speed, peaking, or deload all map to different bands. The percentage-of-1RM calculator carries both the NSCA traditional band (the field default for ~30 years) and the ACSM 2026 Position Stand update; the methodology page covers where the two framings differ. For programming around a 1RM (training max, peaking attempts, RIR-driven percentages), the calculator's output and the engine's reliability band are the two numbers worth tracking.