Nonorthogonal analysis of variance has recieved considerable attention in recent years. It arises when there are unequal and disproportionate cell sizes in a factorial design. The unequal (or empty) cells result in effects which are correlated and nonindependent. Thus, tests of main effects and interactions are not independent and are confounded with one another. Several least squares approaches exist for nonorthogonal data. Three that are frequently used are Overall and Spiegel's Method 1, 2, and 3. Method 1 involves the testing of each main effect and interaction after adjusting for or partialing out every other effect in the analysis. Method 2 involves adjusting each effect for every other effect at the same or lower level. Method 3, the hierarchical method, involves testing effects in an order with only the effects of preceding variables removed. A number of research papers have suggested the proper use of these methods: method 3 is appropriate when testing a hierarchical series of effects; method 2 is appropriate when interactions do not exist; and method 1 is of general utility when proper dummy or effect coding is done, although it may result in nonadditive effects. Tabular data, 2 notes, and 28 references are included.