Objective To investigate the diagnostic value of random forest(RF)model based on CT images and clinical data for preoperative T staging of colorectal cancer. Methods Four hundred and fifty patients with colorectal cancer who were pathologically confirmed by surgery and underwent preoperative CT examinationinthe first hospital of Jilin university from January 2016 to July 2016 were included retrospectively(Stage≤T2,T3,and T4 each has 150 cases).According to the ratio of 2:1,the patients were divided into training set(300 cases)and test set(150 cases,stage ≤T2,T3,and T4 each has 50 cases)by computer random software. Each of 450 patients had one lesion. All the patients underwent preoperative abdominal and pelvic contrast-enhanced CT scan.The clinical,imaging and pathological data[gender,age, carcinoembryonic antigen (CEA) level, carbohydrate antigen 19-9 (CA19-9) expression, intestinal wall deformation, maximum diameter of tumorand thickness of intestinal wall, location, enhancement homogeneity and enhancement rate]of these patients were collected.The correlation between the collected factors and pathological T staging was analyzed by Spearman correlation analysis.The preoperative staging model of colorectal cancer was established by RF algorithm in the training set.Two kinds of methods(model and traditional method)were used to diagnose T stage of the patients in the test set.The accuracy of the two methods was calculated by postoperative pathological staging as the gold standard.The consistency test was used to evaluate the consistency of the RF model results with the pathological results. Results T-staging was positively correlated with CEA, CA19-9, intestinal wall deformation, tumor size and thickness of intestinal wall(r=0.449,0.291,0.624,0.573,0.386;P<0.05).Age,location,enhancement homogeneity and enhancement rate were slightly negatively correlated with T-staging(r=-0.115,-0.245,-0.120 and-0.339;P<0.05).The predictive results of the model in≤T2,T3,and T4 stage cancers were moderately and highly consistent with the standard of pathology,and the Kappa value were 0.769,0.615 and 0.800,respectively.The total accuracy rate of the model andthe traditional method are 80.7%(121/150)and 54.0%(81/150). Conclusion Application of random forest model based on multi-slice spiral CT images and clinical data can improve the diagnostic efficacy of preoperative T stage of colorectal cancer.