Small area estimation based on area level models typically assumes that sampling error variances for the direct survey small area estimates are known. In practice we use estimates of the sampling error variances, and these can contain substantial error. This suggests modeling the sampling variances to improve them and to quantify ed’ects of their estimation error on small area inferences. We review papers that have attempted to address these issues. We then provide some results on the latter issue, showing, in a simple framework, how error in estimating sampling variances can ad’ect the accuracy of small area predictions and lead to bias in stated mean squared errors.