The validity of the Computer Science and Applications, Inc. (CSA) accelerometer in assessing physical activity was assessed during treadmill walking and running at three different grades. Energy expenditure (EE) served as the criterion measure. CSA data were compared to data collected with the Caltrac accelerometer. Both accelerometers were sensitive to changes in treadmill speed, but neither discriminated changes in treadmill grade. Caltrac and CSA activity counts were significantly and similarly correlated with EE (r = 0.66-0.82), relative VO2 (r = 0.77-0.89), heart rate (r = 0.66-0.80), treadmill speed (r = 0.82-0.92), and with each other (r = 0.77-0.82). CSA data were used to develop models to predict EE (kcal.min-1). Cross-validation resulted in a mean difference between actual and predicted EE of 0.02 kcal.min-1 (SEE = 0.85 kcal.min-1). The range of individual differences in the validation group was large for both the CSA model (-2.86 to +3.86 kcal.min-1) and Caltrac (-4.17 to +2.04 kcal.min-1). It is concluded that the CSA and Caltrac accelerometers have similar validity and that either instrument can be used to estimate EE of groups.