In this paper an object-based selection method is proposed on how to select the best products from lots of numerical forecasts. This method firstly gives single object score with different weight means of TS score, center of gravity score, area size score, and location shape score. On this basis, the regional forecast overall evaluation and selection are conducted. Batch comparative experiments of 3 months and cases studies with local scattered rainfall and large-scale rainfall weather processes are conducted. The results show that the object-based selection result is reasonable. The weight of every score item of single object has great impact on the selection result. Increasing the weight of TS score (exceeding 0.4) can improve the selection result. It is the key factor to calculate regional forecast score based on every object area size weight score in the forecast region. The object-based selection method has remarkable advantages for local scattered rainfall and it can overcome TS score’s shortcoming. Usually the object-based selection result is more reasonable because it includes the shape, center of gravity and area scores of rainfall forecast.