Students of a teacher training program (N=124) were asked to generate, unaided, a binary sequence of 40 symbols appearing to be as random as possible. It was found that the average probability of alternation, P(A), was 0.61, which agrees with previous experiments described in the literature, and that the students tend to underestimate the occurrence of runs with four or more equal symbols. When comparing to random sequences I argue that only asking for sequences of length 40 will overestimate P(A).