Multivariate normality is required for some statistical tests. This paper explores the implications of violating the assumption of multivariate normality and illustrates a graphical procedure for evaluating multivariate normality. The logic for using the multivariate bootstrap is presented. The multivariate bootstrap can be used when distribution assumptions are not met, or for descriptive purposes in all cases. Multivariate bootstrap logic is illustrated for the canonical correlation case. From a practical point of view, computer automation is required for the bootstrap approach. Various types of software for conducting multivariate bootstrap analyses are described. An appendix presents computer program output for the data. (Contains 1 figure, 4 tables, and 29 references.) (Author/SLD)