The accuracy of technical efficiency measures is important given the interest in such measures in policy discussions. In recent years the use of stochastic frontiers has become popular for estimating technical inefficiency, but estimated inefficiencies are sensitive to specification errors. One source of such errors is heteroscedasticity. This paper addresses this issue by extending the Hadri (1999) correction for heteroscedasticity to stochastic production frontiers and to panel data. It is argued that heteroscedasticity within an estimation can have a significant effect on results, and that correcting for heteroscedasticity yields more accurate measures of technical inefficiency. Using panel data on cereal farms, it is found that the usual technical efficiency measures used in stochastic production frontiers are significantly sensitive to the extended correction for heteroscedasticity.