Purpose: To develop a prediction model that combined magnetic resonance images (MRI)-based radiomics features with clinical factors to predict recurrence-free survival (RFS) of hepatocellular carcinoma (HCC) patients treated with surgical resection. Methods: HCC patients treated with surgical resection (n = 153) were randomly divided into training (n = 107) and validation (n = 46) datasets. The volumes of interest were manually outlined around the lesion and additional 2 mm and 5 mm peritumoral areas were created with automated dilatation in MRI to extract tumoral (T) and peritumoral (PT) radiomics features. The radiomics models were constructed using least absolute shrinkage and selection operator Cox regression. The combined model incorporated clinical factors and radiomics features using multivariable Cox regression based on the Akaike information criterion principle. Predictive performance of different models were evaluated by receiver operating characteristic (ROC) curves, decision curves, and calibration curves. Results: Among the radiomics models, similar performance was observed in the 2 mm and 5 mm PT models (C-index both 0.657), which were better than the T model or T + PT model (C-index 0.607 and 0.641, respectively) in the validation dataset, whereas the model combined with the three identified clinical risk factors showed the best performance (C-index 0.725). Results of the ROC curves, decision curves, and the calibration curves indicated that the combined model and the derived nomogram had better prediction performance, greater clinical benefits, and fair calibration efficiency. Conclusion: The prediction model that combined MRI radiomics signatures with clinical factors can effectively predict the prognosis of patients with HCC treated with surgical resection. © 2021, The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature.