In this study, a new knowledge discovery and data mining-based technique has been proposed for garment size selection. It could be split into four sequential parts principally, involving data preparation, data preprocessing, fit models setting, and size selection. Two cases of mass customization, representing the top and bottom garments respectively, were utilized to expounding the implementation of the presented approach. After data preprocessing, key body dimensions were identified using the hierarchical clustering algorithm. Next, the enumeration algorithm was utilized by listing all the possible values while computing the distance between the target population and the fit models. Afterwards, an improved K-means clustering algorithm and support vector machine (SVM) method were utilized to size selection, respectively. Eventually, the SVM-based solution was considered as the optimal solution after being evaluated by the aggregate loss of fit, number of poor fit, accommodation rate of ideal fit, and number of sizes employed. The experimental results demonstrate that the present approach is a low-cost and high-effective improvement for size selection by exploiting the potentials of the existing sizing system, without creating new sizing systems. Moreover, the proposed approach can easily be applied to any type of garment in a flexible manner.