According to data collected from Fujian statistical yearbook in 2013, using GM (1, 1) model and a three-layer BP neural network structure model build by MATLAB to predict population mortality of Fujian province respectively. Result shows that BP neural network has an better fitting result in population mortality predicting than GM (1, 1) model, which minimum relative error is 1.05%, maximum relative error is 3.75%, and average relative error is 2.54%, this fitting result shows that model is reliable. BP neural network,which contains an nonlinear structure system, has higher prediction accuracy and stronger feasibility characteristics，so BP neural network could be used to forecast future population scale, and providing reference of economic and social development strategy planning and adjusting structure of population , making reasonable population decisions for Fujian province.