This paper studies the decomposition of the time reversal operator (DORT, by the French acronym) technique for microwave breast lesion classification. We apply the finitedifference time-domain (FDTD) method to a realistic numerical breast phantom where lesion-like targets are artificially introduced, and obtain the multistatic data matrix (MDM) for a particular antenna array configuration. Then, the singular value decomposition (SVD) of this matrix is derived for different targets, which represent malignant and benign lesions. We show that the singular value spectrum can assist in classifying these targets as malignant or benign, especially in the case where contrast-enhanced agents can be employed to allow the analysis of differential backscatter data.