The measurement of the impact of IT spillovers on productivity is an important emerging area of research. Studies of IT spillovers often adopt a ‘production function’ approach commonly used for measuring R&D spillovers, in which an external pool of IT investment is modeled using weighted measures of the IT investments of other firms, industries, or countries. We show that when using this approach, measurement error in a firm’s own IT inputs can create a significant upward bias on the estimated social returns to IT investment. This problem is particularly severe for estimating IT spillovers due to the high levels of measurement error in most available IT data. This bias can be addressed by using instrumental variable techniques to correct the measurement error in a firm’s own IT inputs. Using panel data on IT investment, we show that measurement error corrected estimates of IT spillovers are 40 to 90% lower than uncorrected estimates. This bias term is increasing in the correlation between the IT pool and firms’ own IT investment. Therefore, when instruments are not available, the use of fine-grained data on transmission paths can be an effective solution because IT spillover channels are more likely than R&D spillover channels to cut across industry boundaries, minimizing the correlation between a firm’s own IT investment and the constructed external IT pool. Implications for researchers, policy makers, and managers are discussed.