High mortality arte due to breast cancer can be reduced if it is diagnosed at the early stage. Clinical Infrared Thermography is a non-contact, non-invasive, non-hazardous technique which is widely accepted as reliable tool for early detection of breast cancer. It maps the heat emitted from the region into thermal patterns called thermographs. Thermographs of a normal person show uniform and symmetrical pattern. On the other hand thermographs of affected persons result in non-uniform and asymmetrical thermal patterns. Interpretation of these thermographs helps in identifying the abnormality and classifying the same. In this paper the impact of learning method on the performance of automatic classifier system based on BPN is studied in terms of Mean Square Error. Levenberg-Marquardt method provides better accuracy with the least Mean Square Error.