Breast cancer could be cured 99.4% if all tumors are detected by the time they reach 0.4 cm. However, since many normal structures in the breast will be of this magnitude, to detect small tumors we need to align two breast images taken some time apart and subtract them. Hence image registration is important for breast imaging. Image registration consists in identifying corresponding points in two images and interpolating between them. The low contrast and noise in breast images permit few control points. The "starbyte" textural transformation is introduced and used to solve the breast image registration problem. A textural value at a pixel is obtained as a bit string using binary comparisons of average grey values over different sectors in the local neighborhood of a pixel. Starbyte transformations can be done in several ways, converting breast images to texture maps with clearly demarcable regions from which accurate control points can be easily identified. The performance of the starbyte registration algorithm is evaluated for six different types of distortions. A manual and an automatic algorithm for identification of control points have been presented in detail with several examples. Limited studies with noise and simulated tumors have also been carried out. A complete registration algorithm is presented which consists of first starbyte-transforming the original images, extracting control points, and performing unwarping using thin plate splines. Mammography has limitations because of its two dimensional nature. Abnormalities are missed primarily because of tissue overlap. We show that direct registration of longitudinal mammograms cannot give us useful results, demonstrating the need for 3D breast imaging. Starbyte registration of pairs of roughly corresponding slices from 3D breast MRIs shows the advantage of 3D. An important feature of starbyte registration is that it can be easily extended to 3D.