The activPAL is a widely-used measure of sedentary time but few studies have evaluated its ability to estimate physical activity intensity. This study determined the accuracy of the algorithm used by the activPAL to predict metabolic equivalents (METs) from cadence and a curvilinear cadence-METs equation individualized for height. Thirty-six healthy adults (25 ± 6 years) completed a progressive walking protocol. Stepping cadence was video recorded and METs were determined via indirect calorimetry. Manually-counted cadence was input into the activPAL and curvilinear equations. The internal activPAL equation overpredicted METs at slower cadences (<120 steps/minute) but underpredicted METs at faster cadences (>120 step/minute) (proportional bias, p < 0.001). Conversely, the curvilinear equation exhibited neither fixed (p = 0.37) nor proportional bias (p = 0.07), and a lower absolute MET difference [0.87 ± 0.65 (range:0.0-3.2) vs. 0.56 ± 0.45 (range:0.0-2.7) METs]. The linear activPAL equation poorly estimates METs from stepping cadence but these inaccuracies may be lessened through the use of an individualized curvilinear equation.