Three person-based computer files were linked to provide a data-set of a random sample of 32,679 Swedes, drawn for interviews regarding perceived health, socioeconomic conditions, and psychoactive drug use. All diagnoses from inpatient psychiatric care in the sample during a 15-year period and the causes of death after the sampling point were combined with the interview responses. Among those admitted for inpatient psychiatric care, substance abuse was an infrequent diagnosis; the majority of schizophrenics and of those with an affective disorder appeared not to medicate regularly; survey non-responders had higher rates of mental disorders than responders. Drug use correlated with both subjective and objective measures of mental ill health. The rate of prescription drug abuse was low. Automated record-linkage is a feasible method to generate hypotheses about mental health in the general population.