In this paper, 316L stainless steel samples are fabricated by selective laser melting technology, and the microstructure of samples are analyzed by scanning electron microscope and optical microscope. The effects of laser power and line energy density (LED) on the upper surface morphology of the sample are studied. With laser power and scanning speed as input, the roughness of upper surface of forming samples are predicted based on genetic algorithm optimized back propagation (GA-BP) neural network, The experimental results show that the LED has a great influence on the surface morphology and forming defects of fabricated samples. When the LED is 240 J/m, the melt track is smooth and continuous. The mean absolute percentage error of the GA-BP neural network prediction model is 6.34%.