We evaluate consequences of some important assumptions ofthe perpetual inventory method of capital stock calculation under geometric depreciation. The data are plant-level panel data from the Norwegian manufacturing statistics, containing independent measures of capital stocks and gross investment flows for two capital types and three industries. First, we look at consequences of choosing different depreciation rates a priori, when we use as benchmark for the level of the capital stocks deflated fire insurance values in a specific year. The choice of depreciation rate is of substantial importance, some values resulting in decreasing, other in increasing capital stocks over time. Second, we attempt to estimate depreciation rates by combining time series on gross investment and fire insurance values for the same period. In our regression models, both systematic and random measurement errors in the fire insurance values and various forms of heterogeneity in the coefficient structure are represented. We conclude that the estimated depreciation rates vary significantly with the specification of the measurement error process and that heterogeneity in this process across plants is important.