In Experiment 1, subjects estimated a) the mean of a random sample of ten scores consisting of nine unknown scores and a known score that was divergent from the population mean; and b) the mean of the nine unknown scores. The modal answer (about 40% of the responses) for both sample means was the population mean. The results extend the work of Tversky and Kahneman by demonstrating that subjects hold a passive, descriptive view of random sampling rather than an active balancing model. This result was explored further in in-depth interviews, wherein subjects solved the problem while explaining their reasoning. The interview data replicated Experiment 1 and further showed (a) that subjects' solutions were fairly stable-- when presented with alternative solutions including the correct one, few subjects changed their answer; (b) little evidence of a balancing mechanism; and (c) that acceptance of both means as 400 is largely a result of the perceived unpredictability of "random samples."