From the last quinquennium, Cloth-Changing Person Re-Identification (CCreID) has been explored thoroughly but still lags in proper augmentation studies for cloth variation. Existing CCreID methods encounter challenges due to inadequate cloth variations in the training set. To address this issue, we propose a novel cloth augmentation strategy, dubbed ClothMix, that enhances the cloth texture diversity in training data by randomly exchanging feature moments (means and variance) between clothing regions of training samples in feature space. Concretely, we divide image features into identity-related (head part) and clothing regions (upper and lower clothes) and randomly mix feature moments within the same clothing regions among batch samples to diversify clothing styles. ClothMix easily integrates with CCreID methods and other augmentation strategies during training. We conducted extensive experiments on cloth-changing benchmarks, including LTCC, PRCC, and VC-Clothes datasets. On evaluation, results illustrate that ClothMix significantly improves the model’s performance in handling cloth-changing tasks.