In regression analysis,there are usually more than one estimation approaches for one model hypothesis.In general,which approach is better for a certain pattern of data is unknown.The determination of estimation approaches depends on distributions of data,and the distribution itself decides which approach is better.If the better approach for the corresponding data is not clear,the less efficient approach might be chosen and expand the variance.We propose an estimating method of nonparametrical combination to decrease the risk,a method to find out an appropriate proportion that help we conclude a more effective method.Besides,the method doesn't require a lot of assumptions on distributions or forms of regression functions,therefore it can be applied in estimations of different kinds of experiment data.Finally we compare the performance between estimation methods by calculating average estimated numbers,variances,and 95% confidence interval coverage percent through computer simulations,and the result is discussd.