This study of psychiatric illness among 4098 patients attending 91 general practitioners compares 2 methods of case identification: 'conspicuous morbidity' by the doctor's own assessments, and 'probable prevalence' by the patients' responses to the General Health Questionnaire (GHQ). In general, the latter gives somewhat higher estimates than the former, but there are wide variations in morbidity between practices. The ability of each general practitioner to detect psychiatric illness was measured by computing Spearman's correlation coefficient between his assessments and the GHQ scores of his patients. The mean correlation coefficient was + 0.36, but the range was very wide (0.09-0.60). The first part of the study deals with various demographic characteristics of the patients themselves which are associated with an increased likelihood of the doctor detecting a psychiatric illness; such factors include unemployment, female sex, and marriages which have ended by separation, divorce or death. The second part of the study examines characteristics of the doctors themselves in an attempt to account for the wide variation between them in their ability to detect psychiatric illness. A research psychiatrist made detailed observations on 2098 interviews carried out by 55 general practitioners. Each doctor's verbal and non-verbal styles were recorded minutely, and in addition various global ratings were made. The doctors completed personality inventories and supplied details of training and professional background. It was possible to account for 67% of the variance of correlation coefficient mainly in terms of 2 dimensions: 'interest and concern' and 'conservatism'. The way in which the doctor interviews his patients is shown to be important, but there are interactions between interview style and the doctor's personality.