Two modifications to standard Aitken estimation of heteroskedastic linear regressions are suggested: (1) different criteria for calculating unknown parameters to yield more efficient estimates to the coefficients when the form of heteroskedasticity is misspecified and (2) a different way of estimating the covariance matrix of the feasible Aitken estimator. A sampling experiment provides evidence that the first adjustment could provide more efficient estimates, though not clearly more so than alternatives. The second adjustment is crucial for reliable inference, even when the form of heteroskedasticity is correctly specified.