Objective To evaluate a score system(Association of Coloproctology of Great Britain and Ireland ACPGBI)in prediction of postoperative mortality from colorectal cancer patients in a Chinese hospital. Methods We analyzed retrospectively 904 patients with histologically confirmed colorectal cancer who had colorectal surgery from 1992 to 2005.There were 525 colonic cancer patients and 379 rectal cancer patients.We divided patients into several groups according to operative urgency(elective or emergency);surgeons(colorectal specialists or other surgeons);cancer location(colon or rectal).According to ACPGBI score we got the prediction.This prediction was compared with the actual mortality;Chi-square test,receiver operator characteristic curve(ROC),Hosmer-Lemeshow goodness-of-fit test were used.Results Observed overall mortality within 30 days after surgery was 1.0%(9/904),and the predicted mortality was 8.3%(75/904).In all the subgroups the predicted momdity wag higher than observed mortality.We found that the actual mortality was higher in an individual subgroup in which the predicted mortality was higher. Conclusions For colorectal cancer patients undergoing a surgery the predicted mortality of ACPGBI score system was higher than the actual mortality in a Chinese hospital.