This article compares the power of the test RESET to that of a number of autocorrelation tests in detecting the errors of omitted variables or incorrect functional form in regression analysis. The autocorrelation tests considered are the Durbin-Watson test and a chi-squared test on the first H autocorrelations in the residual vector. Monte Carlo experiments reveal that the RESET test is the most powerful test for detecting specification errors and is robust to autocorrelation.