Due to manufacturing process immaturity, the parameter dispersion of Silicon Carbide(SiC) MOSFET is severe, which has risks of current imbalances and oscillation in parallel applications. In addition, as the bipolar degradation is gradually solved, SiC MOSFET body diodes are used to replace anti-parallel diodes to conduct at deadtime. Therefore, the screening method needs to consider the current sharing capability in dynamic, static, and reverse processes at the same time. This article analyzes the body effect of SiC MOSFET and the feasibility of taking output curves at multiple gate voltages as screening objects. Then, the concept of multivariate statistical analysis is introduced to quantify the similarity of output curves into distance parameters, and an automatic screening method based on hierarchical clustering is proposed. The validity that the distance parameters at room temperature can be used in a wide temperature range(25-150 °C) is evaluated. Finally, a double pulse test platform with a highly symmetrical circuit is built to test the performance between the proposed method and traditional screening methods. As a result, the error between paralleled devices selected by the proposed method is less than 5% in dynamic and reverse processes and 2% in static processes simultaneously.