In a training programme to improve the accuracy with which family doctors rate psychiatric disturbances in their patients an agreement coefficient (kappa) was calculated for each of 45 trainees between their ratings of psychiatric disturbances and the symptom levels of their patients as reported on the General Health Questionnaire (GHQ-28). The 24 trainees with the lowest kappa cofficients were randomly assigned to an index or a control group, and the index group received four sessions of individual instruction, based on videotaped feedback of their own interviews, aimed at modifying their interview style. After the training sessions the index group had significantly improved the accuracy of their assessments.