In the educational literature, responses to surveys commonly serve as the source of data for many empirical articles. Whenever a survey is used as a source of data, the response rate can greatly affect the potential generalizability of the findings. Using Monte Carlo methods, this study examined the effects on sample estimates of the population mean and standard deviation for 3 levels of effect size differences between the responders and nonresponders (0.0, 0.25, and 0.50). Two data sets were used: 400 normally distributed random values and 200 responses to an item on a Likert-type scale. The number of replications for each condition was 5,000. The proportion of population values contained within a 95% confidence interval of the sample estimates was then calculated with respect to the mean and standard deviation. For the 0.0 effect size conditions, all response rate levels produced expected proportions of samples containing the population values. Increased effect size differences combined with reduced response rate levels resulted in biased estimates, particularly for the mean. Although return rates of 70% have been recommended as adequate, response rates of at least 90% may be needed if moderate effect size differences are suspected between responders and nonresponders. (Contains 4 tables and 14 references.) (Author/SLD)