This paper deals with the estimation of the impact of technology spillovers on productivity at the firm level. Panel data for American manufacturing firms on sales, physical capital inputs, employment and R&D investments are linked to R&D data by industry. The latter data are used to construct four different sets of `indirect' R&D stocks, representing technology obtained through spillovers. The differences between two distinct kinds of spillovers are stressed. Cointegration analysis is introduced into production function estimation. Spillovers are found to have significant positive effects on productivity, although their magnitudes differ between high-tech, medium-tech and low-tech firms.