It is argued here that the choice of the appropriate method for calculating least squares analysis of variance with unequal sample sizes depends upon the question the experimenter wants to answer about the data. The different questions reflect different null hypotheses. An example is presented using two alternative methods. (Author/BW)