One of the important problems in the analysis of variance is to determine the proper denominators for F statistics. The test statistics is a ratio of mean squares that is chosen such that the components of the expected value of the numerator mean square is more one component than the expected value of the denominator mean square. This reseach introduced the rules that produced the expected mean squares for any model and some examples as follows: one way randomized block design, two way completely randomized design, two way randomized block design, two way fixed model design, completely randomized hierachical design, and latin square design. The possible solutions of no exact test statistics for some effect in the designs were discussed also.