There isn't one formula that's best for everyone — there's a best formula for what you can actually measure. Here's how the three common equations compare and when each one wins.
For most people without a body-fat measurement, Mifflin-St Jeor is the most accurate. When researchers compared the common equations against resting energy measured by indirect calorimetry, Mifflin-St Jeor predicted it within 10% more often than the alternatives — which is why it's the modern default here and on most reputable calculators.
Mifflin-St Jeor (1990). Uses weight, height, age, and sex. The best general-purpose choice; no body-fat figure needed.
Revised Harris-Benedict (1984). The same inputs, derived from an older population. Tends to read slightly higher and is marginally less accurate in modern comparisons — useful mainly as a reference point. See the head-to-head on Mifflin vs Harris-Benedict.
Katch-McArdle. Driven by lean body mass, so it needs a body-fat percentage. When that measurement is reliable, it can beat the height/weight formulas for people with atypical body composition — very lean, very muscular, or higher body fat. Details on the Katch-McArdle page.
The accuracy ranking flips on one question: do you have a trustworthy body-fat measurement? If yes, Katch-McArdle's lean-mass basis can be the most accurate for you. If no — and most people don't have a good one — a guessed body-fat percentage produces a guessed BMR, and Mifflin-St Jeor's height/weight inputs are the safer, more accurate bet.
Every equation is a population estimate. None can see your genetics, hormones, or non-exercise movement, so any output can be off by a couple hundred calories for an individual. That's why the durable approach is to start with Mifflin-St Jeor and then calibrate against your real weight trend — which the tracker does. Full sourcing is on the methodology page.
Mifflin-St Jeor for most people without a body-fat measurement — a systematic review found it predicts resting energy within 10% more often than the alternatives. Katch-McArdle can be more accurate if you have a reliable body-fat percentage.
Only when its body-fat input is accurate. With a reliable measurement it can beat Mifflin-St Jeor for atypical body composition; with a guessed one, Mifflin-St Jeor is safer.
They use coefficients from different populations and inputs. For most people the spread is small. No formula captures individual genetics or activity, so treat any result as an estimate to calibrate against reality.