Of all neuroleptic-naive, acutely psychotic subjects admitted to hospital over a 2-year period (n = 62), 27 participated in a neuroendocrine study and 35 did not participate (51% refused consent, 19% were incapable of consent and 31% started neuroleptics immediately). However, all nonparticipants agreed to psychopathological evaluation, thus allowing comparison between participants and nonparticipants. The 2 groups were similar in most respects, except that more nonparticipants were hostile. Among subjects with schizophrenia, 47% of nonparticipants had the paranoid subtype vs 8% of participants. There was also a trend towards longer illness duration in nonparticipants. These results underline the need for neurobiological studies of psychosis to consider sample bias as a confounding variable.