To solve the pvoblem that the masses on the and have the fuzzy boundaries

the paper presents a method gram equalization theory firstly

and extract four features of mammograms exhibit poor contrast to the background to enhance the image by wavelet transform and histoevery pixel. According to the principle that artificial neural network can implement classification through training

the paper presents the method to classify every pixel to the masses or not by BP neural network to achieve the detection of the masses. The method has simple operation

and is evaluated by 30 mammograms images in the MIAS

and the effectiveness is 83%.