To improve detectability in abnormal shadow detection in breast images.

In an abnormal shadow detector, whether or not an abnormal shadow candidate extracted in the breast image P is an abnormal shadow is discriminated on the basis of the combination K of a prescribed feature amounts for the candidate and the candidate discriminated as being the abnormal shadow is detected as the abnormal shadow. The abnormal shadow detector is provided with a breast image classification means 140 for classifying the breast image P into a plurality of kinds corresponding to the size of a mammary gland region in the breast image P and a feature amount combination setting means 150 for setting the optimum combination for the combination K corresponding to the classified kind.