Abstract: | The analysis-of-variance tests for hypotheses on random effects in regular linear models are considered. Conditions are given for these tests to be uniformly most powerful unbiased or uniformly most powerful invariant unbiased. An example shows that the difference between these conditions can be serious. |