The goal of computer-aided diagnosis (CAD) is to help reduce errors by radiologists when interpreting medical images. Two types of errors can be made. The first is overlooking an important finding, such as a cancer. The second is misinterpreting a finding, such as believing a malignant lesion is benign. In breast cancer detection, these can be large with a miss detection rate as high as 40% and having unnecessary biopsy performed in 2 out of every 3 women biopsied. There are two different types of CAD schemes being developed: computer-aided detection (CADe), where suspicious lesions are identified by the computer, and computer-aided diagnosis (CADx) where the likelihood of malignancy of an identified lesion is determined by the computer. In this talk, I will describe the basic principles and techniques used to develop CADe and CADx schemes and I will present laboratory and clinical results showing the potential of CAD. I will emphasize how our knowledge of the physics of image acquisition and our knowledge of how radiologists approach their clinical task are critical in developing and implementing CAD schemes