Background: Clear cell sarcoma of the kidney (CCSK) is a rare but the second common renal malignant tumor mimicking Wilms’ tumor. Radiomics is helpful for differentiating CCSK from Wilms’ tumor preoperatively through analyzing the pixel distribution of lesions on medical images quantitatively. Procedure: In this study, the regions of interest (ROIs) of lesions were delineated on corticomedullary phase (CMP) and nephrographic phase (NP) images to extract radiomics features. Dimensionality reduction and Logistic Regression (LR) algorithm were used to construct the classification models. The area under the receiver operator characteristic curve (AUC), sensitivity and specificity were calculated for evaluation, and Delong test was used to compare the performance of the most meaningful features and LR models. Results: Lower skewness was observed in Wilms’ tumor, and higher skewness in CCSK. Skewness transformed by exponential and squareroot filters from NP images achieved moderate to good diagnostic performance for CCSK with AUCs of 0.707 (95%CI: 0.573, 0.840) and 0.705 (95%CI: 0.572, 0.839) in the training set, and 0.818 (95%CI: 0.608, 1.000) and 0.803 (95%CI: 0.585, 1.000) in the validation set, respectively. Delong test showed no significant difference between LR model, exponential-skewness and squareroot-skewness based on NP images in both training and validation sets. Conclusion: Skewness from nephrographic phase at exponential and squareroot filters is helpful to discriminate between CCSK and Wilms’ tumor in children, and higher skewness on NP images may be a potential imaging biomarker for diagnosing CCSK from Wilms’ tumor.