PURPOSE: To determine whether breast cancers missed at screening mammography have distinguishing characteristics from those of detected cancers. MATERIALS AND METHODS: The mammograms of 146 women with mammographically identifiable breast cancer were viewed independently by two radiologists who were blinded as to whether the cancer had been missed or detected (group 1 lesions, missed cancers; group 2 lesions, detected cancers) at screening. The mammographic lesions were characterized as to location, size, density, type, and visibility on two views. RESULTS: A significant difference between missed and detected cancers was found for diameter (P = .03), number of views (P < .0017), and density (P = .0007). Stepwise multivariable logistic regression showed that density (P = .01) and the number of views (P = .03) but not diameter (P = .27) were independently significant in distinguishing the groups. No statistically significant difference was found between the two groups for lesion type (P = .32 for reader 1 and P = .27 for reader 2) or location (P = .86 for reader 1 and P > .96 for reader 2). CONCLUSION: Missed cancers were statistically significantly lower in density and more often seen on only one of two views than detected cancers.