Most schizophrenia research is undertaken on clinical samples in current contact with mental health services. It is not clear to what extent such samples are representative of the population of people with schizophrenia or whether they differ significantly, for example, from those who are being treated predominantly in primary care settings or who are recruited from non-clinical sources. Data from a volunteer schizophrenia research register and two recent Australian studies are reported, the Low Prevalence (psychotic) Disorders Study and an associated study of schizophrenia in general practice, in which all participants completed the same clinical assessment interview. Participants meeting criteria for schizophrenia or schizoaffective disorder were classified according to their source of recruitment: volunteer research register (n=128), general practice (n=123), community (n=236) or public inpatient (n=178) mental health services. Hierarchical discriminant function analyses revealed significant differences between these recruitment sources with respect to illness-onset factors, relationship and support factors, current functioning and course of illness. A severity/functioning gradient was observed across the four recruitment sources, possibly reflecting a spectrum of neurobiological impairment from good to poor prognosis. The implications of these findings for sampling strategies in schizophrenia research are discussed.