An investigation was conducted of the characteristics of two estimation procedures and corresponding test statistics used in the analysis of completely randomized factorial experiments when observations are lost at random. For one estimator, contrast coefficients for cell means did not involve the cell frequencies. For the other, contrast coefficients were based on the ratios of cell frequencies to marginal totals. Given the usual normal-theory assumptions and assuming that data lost was unrelated to the treatments, the mariginal distributions of estimators and test statistics (across all admissible cell frequencies) were investigated. Marginal means, variances, and probabilities of Type I and Type II errors are presented. (Author)