Past occupational preference is used to estimate the gender pay gap. The use of predetermined variables in a reduced-form wage equation avoids the bias caused by using variables that are correlated with the random error. Using a gender coefficient, the potential discriminatory gap is about 11.5% when past occupational preference is included. Decomposition yields an estimate of 10.5% when past occupational preference is included. In both cases, the discriminatory gap is close to that obtained when actual occupation is included. This suggests public policy directed toward reducing hiring discrimination by gender might be misdirected.