The performance of the General Health Questionnaire as a case detector was assessed in 75 patients receiving chemotherapy for early breast cancer. The questionnaire and its subscales were found to be valid when compared with observer ratings. The misclassification rate of the 60-item version of the questionnaire was lowest when the standard threshold score for clinical morbidity was doubled to 24. It was lowest for the 28-item version when the threshold score was doubled to 10. Suggested threshold scores for clinical morbidity for the subscales are: anxiety and insomnia subscale 6/7; severe depression subscale 2/3; somatic symptoms subscale 6/7; and social dysfunction subscale 7/8. However, lower thresholds would be appropriate if the questionnaire were used as screening instrument prior to interview by a trained observer.