Five patients with breast cancer were referred for mammography at first presentation. In each case, a negative report accepted at face value resulted in a delay of definitive treatment of two to 36 months. When the patients presented again, their tumours were either locally advanced or three or more axillary nodes were involved. We analysed the probabilities which affect the accuracy of any mammogram report. All the potential clinical situations examined yielded an unacceptably high probability of error, which precluded the use of mammography as a non-invasive technique for diagnosing a palpable tumour and excluding the presence of a carcinoma.