A statistical model is presented for analyzing the results of group conversation tests in English, developed in a Finnish university study from 1977 to 1981. The model is illustrated with the findings from the study. In this study, estimates of percentile curves for the number of errors are of greater interest than the mean regression line. It was found that for each examinee both the distributions of total error frequency and the relative number of errors depended on the frequency of the examinee's utterances. Although talkative examinees made more errors than those who rarely spoke, proneness to errors decreased as the frequency of utterances increased, and talkative examinees performed better in this test. Male students were found to make more errors than females, but with almost no difference in the relative number of errors of the two sexes. A number of statistical tables are included in the text. (MSE)