Fourteen novel multiplication problems (of the form "a times b equals") were presented on each of five consecutive days to nine third grade classes divided into three groups. Immediate knowledge of results was provided for group one, delayed for group two, and no knowledge of results for group three. Residual scores, obtained by taking the difference between the number of novel problems answered correctly and the predicted number using stepwise multiple regression, were used as the dependent variables. These residual scores were analyzed using a four factor mixed model analyses of variance, with subjects nested within classes which in turn were nested within treatments. The fourth factor was a repeated measures factor (acquisition and retention tests for analyses one, treatment days for analysis two, and performance on non-novel problems for analysis three). No significant treatment effects were found. Significant differences among classes were observed, pointing out clearly how this source of variation, which has historically been implicitly pooled by investigators unfamiliar with statistical nesting, can affect scores. (Author/FL)