Abstract: Objective: To evaluate the predictive value of preoperative enhanced CT radiomic feature models in comparison with clinical-pathological feature models for disease-free survival (DFS) in patients with non-metastatic grades 2-3 clear cell renal cell carcinoma (ccRCC). Methods: A retrospective analysis was conducted on 315 patients with non-metastatic ccRCC who underwent surgical treatment and were pathologically graded as grades 2-3 between January 2013 and December 2020. Preoperative enhanced CT images, clinical-pathological information, and DFS data were collected. The region of interest (ROI) of the lesion was delineated using ITK-SNAP, and radiomic features were extracted using Python. Patients' radiomic scores were calculated using the least absolute shrinkage and selection operator (LASSO) and Cox regression analysis. Clinical-pathological feature models, radiomic models (corticomedullary phase, parenchymal phase, corticomedullary + parenchymal phase), and combined radiomic and clinical-pathological feature models were constructed to predict patients' DFS efficacy. Results: When predicting DFS in non-metastatic grades 2-3 ccRCC patients, the combined radiomic model of corticomedullary + parenchymal phase demonstrated superior predictive efficacy (C-index: training set 0.848, validation set 0.754) compared to single-phase radiomic models (corticomedullary phase C-index: training set 0.832, validation set 0.701; parenchymal phase C-index: training set 0.842, validation set 0.720). However, the combined model incorporating both radiomic and clinical-pathological features exhibited the highest predictive efficacy (C-index: training set 0.857, validation set 0.832). Conclusion: The model combining radiomic features extracted from preoperative enhanced CT images of corticomedullary and parenchymal phases with clinicopathological features assists in predicting postoperative DFS in non-metastatic grades 2-3 ccRCC.