Why dumbbell math is humbler
The 1RM prediction formulas (Epley, Brzycki, Lombardi, O'Conner) all return weight × f(reps) — they don't know which implement you used. What changes between barbell and dumbbells is everything the math doesn't see: each arm stabilizes its own load (no shared bar), the dumbbells can travel a longer ROM at the bottom, the bench may flex differently under uneven load, the rack only stocks every 2.5 kg or 5 lb, and pressing technique tolerates wider grip and path variation under dumbbells than under a bar.
Each of those differences shifts the relationship between submax-rep performance with dumbbells vs absolute strength on a barbell. The calculator's SECONDARY barbell-equivalent range (1.05–1.25×) is wide because the literature can't pin a single conversion factor — and a narrow false-precise number would be worse than an honest wide one.
A worked example — 30 kg per DB × 5 reps
Run 30 kg per dumbbell × 5 reps. Average across the four formulas:
- Per-dumbbell 1RM ≈ 34.4 kg (cross-formula average, HIGH reliability ±2%).
- Total external load ≈ 68.8 kg (per-dumbbell × 2).
- Rough barbell-equivalent (low end, 1.05×) ≈ 72.2 kg.
- Rough barbell-equivalent (high end, 1.25×) ≈ 86.0 kg.
That barbell range (~72–86 kg) is wide because where you actually land depends on your stability skill, your ROM tolerance, and how well your technique transfers across implements. A first barbell bench session shouldn't open at 86 kg × 5 — start at the low end (or below) and re-estimate from real barbell submax reps once you have a few sessions of bar-path practice.
Frequently asked
Why does the calculator give a per-dumbbell number AND a 'total external load'?
The per-dumbbell number is what you grab off the rack — the practical answer to 'what dumbbells do I order for a heavy single?'. The total external load is the sum of both (per-dumbbell × 2) and is the right number for comparing to a barbell load conceptually. Both are surfaced because both answer different real-world questions: ordering dumbbells vs comparing intensity across implements.
Why is the barbell-equivalent range so wide (1.05–1.25×)?
Because there is no consensus-published dumbbell-to-barbell conversion factor, and the academic literature on this is thin. The 1.05–1.25× range reflects practitioner consensus that the same submax-rep performance with dumbbells corresponds to a somewhat higher external load with a barbell — but how much higher depends on five things the math doesn't see (stability demand, ROM differential, bench angle, dumbbell increment granularity, technique drift). A wide honest range is more useful than a narrow false-precise one. The range is gated behind a SHOW toggle so it doesn't overshadow the per-dumbbell answer.
Can I use the barbell-equivalent number to set my barbell working weight?
Cautiously. Treat it as a starting point, not a target. If your dumbbell bench estimate suggests a 100–120 kg barbell range, your first session under a bar should start meaningfully below the low end (say 80 kg × 5) and work up — not jump straight to 100. Stability and bar-path skill differ enough between implements that the carryover isn't 1:1. Re-estimate from your barbell submax reps once you have a few sessions under the bar.
How are dumbbells different mathematically?
The 1RM formulas (Epley/Brzycki/Lombardi/O'Conner) work the same way — weight × f(reps). What changes is what 'weight' means: with a barbell it's the total load on the bar, with dumbbells it's the per-dumbbell weight. The math doesn't know whether you're stabilizing one bar or two independent implements. The caveats list (visible behind the SHOW toggle) covers what the math can't see — stability demand, ROM, angle, increments, and technique.
Should I use the same reliability rules as barbell?
Yes. ≤5 reps for HIGH reliability (~±2% StrengthMath band, informed by LeSuer 1997), 6–10 MEDIUM (~±5%), past 10 NOISY. The reliability bands are about the rep-count math, not the implement. A 5-rep dumbbell estimate has the same per-formula prediction error as a 5-rep barbell estimate; it's the cross-implement comparison (per-dumbbell vs barbell) that gets noisy, not the per-implement estimate itself.
What I'd do next
- Estimate barbell bench from barbell reps
DB-to-barbell conversion is a wide range, not a number. Test the implement you’ll actually program with.
- Why the conversion is a range
The 1.05-1.25× spread, what drives it, and when DB-strong / barbell-rusty profiles show up.
- Build working weights off the per-DB number
Use the per-dumbbell answer, not the barbell-equivalent estimate. Less guessing, more reproducible programming.
Also in this cluster
By Jimmy L Wu. PRIMARY (per-dumbbell + total external load) and SECONDARY (barbell-equivalent range) outputs are deliberately asymmetric — the per-dumbbell number is what you grab off the rack; the barbell range is a wide ballpark, not a conversion. The 1.05–1.25× factor is StrengthMath methodology — practitioner-consensus framing, not a published academic conversion. Engine logic in lib/strength/oneRepMax.ts. Not medical advice — for max attempts, work with a qualified strength coach.