With a sample of 531 children measured on 79 variables, a series of discriminant function predictions of school placement were computed to compare the results of selected alternative procedures for treating missing data in multivariate data analyses. Procedures included elimination of cases with missing data as well as methods for substituting estimated values. While all procedures had some drawbacks, replacement of missing data with randomly drawn values from the available pool of study data resulted in least reduction in the multiple correlation while still maintaining a suitable sample size. Advantages and disadvantages of alternative procedures were discussed. (Author/DEP)