Laser welding is a complicated process,and quantitative analysis of this process is quite difficult. A non parametric statistical method of light intensity distribution modeling,based on normalized radial basis function neural network,is proposed to predict the spatiotemporal dynamics of surface optical activity in the laser welding process. This neural network adopts Gaussian function as radial basis function. A quantitative evaluation method for light intensity distribution of modeling quality is proposed. Parameters are optimized according to this evaluation method. Comparison of predicted images and testing images exhibits a good resemblance.