This paper develops a new technique for measuring the effect of computer usage on U.S. productivity growth. Standard National Income and Product Accounts (NIPA) measures of the computer capital stock, which are constructed by weighting past investments according to a schedule for economic depreciation (the rate at which capital loses value as it ages), are shown to be inappropriate for growth accounting because they do not capture the effect of a unit of computer capital on productivity. This is due to technological obsolescence: machines that are still productive are retired because they are no longer near the technological frontier, and anticipation of retirement affects economic depreciation. Using a model that incorporates obsolescence, alternative stocks are developed that imply a larger computer-usage effect. This effect, together with the direct effect of increased productivity in the computer-producing sector, accounted for the improvement in U.S. productivity growth over 1996-1998 relative to the previous twenty years.