OBJECTIVE: Since patients being treated for bulimia nervosa constitute only a minority of persons with the disorder, the cases seen in clinics may be subject to sampling bias. The aim of this study was to investigate sampling bias as it affects secondary referrals for bulimia nervosa. METHOD: The personal and family characteristics of a consecutive series of 60 women with secondary referrals for bulimia nervosa (clinic subjects) were compare with those of 83 subjects with bulimia who were recruited directly from the community. Most of the data were collected by interview. RESULTS: The demographic characteristics of the two groups were similar. The clinic subjects had a more severe eating disorder and much greater impairment of social functioning. There was no difference between the groups in duration of the eating disorder or level of general psychiatric disturbance. The community subjects were heavier and had stronger family histories of obesity. CONCLUSIONS: There is sampling bias among secondary referrals for bulimia nervosa. The relative absence of persons prone to obesity among secondary subjects is important, since there is evidence that vulnerability to obesity is a poor prognostic feature as well as being a risk factor for the development of bulimia nervosa. The greater social impairment among the clinic subjects is suggestive of greater personality disturbance in this group. Caution is warranted when generalizing from clinic cases to the disorder as a whole.