Since January 1975 a population-based screening programme for the early detection of breast cancer has been carried out in the city of Nijmegen. During five interscreening periods of 2 years each a total of 158 so-called interval cancers were diagnosed. Careful revision of all screening and diagnostic mammograms was executed. Of all interval cancers 26% were 'missed' at the previous screening examination (due to technical or observer error), 16% were radiographically occult at the time of diagnosis and 58% were 'true' interval cancers. Interval cancers were regarded as 'true' when an obvious lesion was observed on the diagnostic mammogram while no suspect signs were seen on the previous screening mammogram. The prevalence of 'missed' cancers did not decline in the course of the screening programme. Radiographically occult tumours were localised, mostly in Wolfe's P2/DY breast parenchyma (83%), 33% were lobular invasive and 25% ductal non-invasive. 'True' interval cancer cases (58%) showed the same overall survival as control breast cancer patients, diagnosed in a non-screening situation. Shortening the screening interval would reduce interval cancer rates and probably further decrease breast cancer mortality in a screened population. However, from the present series of interval cancers 63% would not have been prevented by an annual screening examination. As regards women under age 50 annual screening would still leave 66% of all interval cancers in this age group undetected. Probably more benefit will be gained by searching for new imaging techniques to reduce numbers of 'missed' cancers and to detect lobular invasive and ductal non-invasive cancers in dense breast parenchyma.