The metabolic cost of steady-state walking is well known; however, across legged animals, most walking bouts are too short to reach steady state. Here, we investigate how bout duration affects the metabolic cost of human walking with varying mechanical power, metabolic intensity and duration. Ten participants walked for 10- to 240-s bouts on a stair climber at 0.20, 0.25 and 0.36 m s-1 and on a treadmill at 1.39 m s-1. Oxygen uptake was time-integrated and divided by bout duration to get bout average uptake (V̇O2(b)). Fitting of oxygen uptake kinetics allowed calculating non-metabolic oxygen exchange during phase-I transient and, hence, non-steady-state metabolic cost (C met(b)) and efficiency. For 240-s bouts, such variables were also calculated at steady state. Across all conditions, shorter bouts had higher V̇O2(b) and C met(b), with proportionally greater non-metabolic oxygen exchange. As the bout duration increased, V̇O2(b), C met(b) and efficiency approached steady-state values. Our findings show that the time-averaged oxygen uptake and metabolic cost are greater for shorter than longer bouts: 30-s bouts consume 20-60% more oxygen than steady-state extrapolations. This is partially explained by the proportionally greater non-metabolic oxygen uptake and leads to lower efficiency for shorter bouts. Inferring metabolic cost from steady state substantially underestimates energy expenditure for short bouts.