To improve the efficiency of constraint solving and the performance of semantic feature modeling system,a new method is presented.It represents all of information in features by the feature semantic representation,enhances the manage- ment mechanism of feature’s elements,separates the local features from product models effectively by semantic faces,and achieves the local constraint solving by creating the provisional cells.This method can not only solve the problem of direct manipulations in a complex model,but also enhance the efficiency of semantic feature CAD system.Experiments on comput- er show that this new method is more adaptable and practicable.