Objective：To determine the optimal preoperative magnetic resonance imaging（MRI）radiomics parameters and compare the values in predicting disease-free survival（DFS）of early-stage cervical cancer patients with clinicopathological factors. Methods：A total of 186 patients with early-stage cervical cancers（preoperative FIGO stage ⅠB-ⅡA），who underwent radical hysterectomy with lymphadenectomy during January 2013 and June 2018，were retrospectively reviewed. Multi-sequence MR data，clinicopathologic data， and DFS data were collected；least - absolute shrinkage and selection operator（LASSO）and the Cox proportional hazard regression model were applied to construct a Rad - score. Clinicopathological models（with significant clinicopathological features），radiomics models（Rad-scores for T1CE，DWI，T2WI，T1CE+DWI，T1CE+T2WI，DWI+T2WI，and T1CE+DWI+T2WI），and combined models （the Rad - score model combined with significant clinicopathological features）were compared using a multivariable Cox model. Results：The T1CE radiomics model was the most stable and optimal model（C-index for training，0.798；for validation，0.758）of all the radiomics models. The radiomics model of T1CE showed higher prognostic performance than the clinicopathological model（C-index for training，0.746；for validation，0.641）. The combined model（histological type of adenocarcinoma，lymph-node metastasis together with Rad-score for T1CE）showed the highest prognostic performance in estimating DFS（C-index for training，0.848；for validation，0.784）. Conclusion：An MRI-derived Rad-score of T1CE combined with a clinicopathological model is optimal in predicting DFS in early-stage cervical cancers.