We have examined variation in the interpretation of xeromammograms among radiologists designated to take part in a Canadian multicenter randomized controlled trial of screening for breast cancer. Radiologists read 100 xeromammograms comprising 10 histologically proved cancers, 40 benign abnormalities, and 50 normal films. Radiologists' opinions differed widely on the frequency of suspected or identified cancer. The diagnostic category "suspicion of cancer" or "cancer" was selected by radiologists for 10-55% of the films, and biopsy or aspiration was recommended for 21 to 53% of patients whose films were examined. Agreement on specific diagnostic categories was greatest for the diagnosis of cancer; agreement was least for the diagnosis of benign abnormalities and intermediate for the diagnosis of normality. Known cancers were in general correctly identified. These results indicate a need for development of methods to reduce observer variation in a interpretation of xeromammograms while preserving diagnostic sensitivity and validity. Results also emphasize the importance of developing strategies to ensure quality control in multicenter trials.