Rothenberg's (1984) Edgeworth size correction is applied to tests of linear hypotheses in the linear regression model with AR(1) errors. Previous simulation findings on the effect of autocorrelation in the regressors on over-rejection (e.g. Park and Mitchell (1980)) are supported when the correction is examined analytically in a special case. A simulation study shows that the correction adjusts the size in the right direction, but still leaves substantial over-rejection when the sample size is small and the original amount of over-rejection is large.