In analyzing a 23factorial experiment, the classical analysis of variance techniques will usually be used. In doing so, one must first assume that the “experimental” errors are independent and follow a normal distribution with common variance. However, in many cases the experimenter does not really know the validity of the assumption. Here, a non-parametric approach to a 23factorial design is introduced. The technique is based primarily on the results of multiple comparisons using rank sums due to 0. J. Dunn (1).