Conventional histologic grading of cystosarcoma phyllodes of the breast has not been entirely successful in the prognosis of recurrence or metastasis. Our study first developed a tumor grade classification based on computerized texture features and then compared the classification to conventional grading of these tumors. Evaluation of the tissue architecture of histologic sections was obtained by measuring nine texture features on an image analysis system. Forty cases of cystosarcoma phyllodes were studied. Each parameter was calculated on subimages of 128 x 128 pixels. This size resulted from a preliminary study that confirmed that the difference between texture primitives depends on the area of subimages. We also compared our series to a panel of 20 extramammary sarcomas. The results show that tissue architecture evaluated by texture analysis allows good discrimination between benign, borderline and malignant cystosarcoma phyllodes. Furthermore, extramammary sarcomas and malignant cystosarcoma phyllodes were discriminated well in most cases.