In order to accurately determine the expression of thymidine phosphorylase (TP) in breast cancer, it is important to develop an appropriate system for immunohistochemical staining evaluation. Three raters, pathologists A-C, independently judged the status of TP in both cancer cells and cancer stromal tissue in 586 breast cancers immunohistochemically. TP expression levels were categorized into four groups (scores 0-3), and scores of 2 and 3 were considered as positive. Interrater and intrarater agreement levels in the judgment of TP expression were analyzed by the kappa statistics. By combining three raters' judgments, we sought for the optimal evaluation system that generates the best concordance among all judgments. The numbers of TP-positive cases were 294 (50.2%), 309 (52.1%) and 344 (58.7%) for cancer cells and were 315 (53.8%), 383 (65.4%) and 397 (67.7%) for cancer stromal tissue in the judgments by pathologists A, B, and C, respectively. There was no correlation of TP immunostaining between cancer cells and cancer stroma tissue. Interrater agreement levels were substantial for cancer cells (kappa = 0.652) and moderate for cancer stroma tissue (kappa = 0.438). Intrarater agreement level for TP judgment by pathologists A, B, or C was substantial to almost perfect for cancer cells (kappa = 0.755-0.880), and for cancer stromal tissue (kappa = 0.620-0.752). The level of TP expression estimated under the majority system and the total scoring system setting a cut-off score between 4 and 5 revealed the highest concordance with the three raters' judgments. To accurately evaluate the TP status in breast cancer tissue, three or more raters are needed and the majority system and total scoring system appear appropriate.