This paper proposes a high-power laser bevel cutting roughness prediction method based on particle swarm optimization radial basis neural network. The 40 kW laser bevel cutting system is used to carry out 30°V- bevel cutting test on Q235 carbon steel with 50 mm thickness; based on the orthogonal test results, the regression prediction model between the laser bevel cutting process parameters and the roughness of the bevel cut surface is established by the radial basis neural network; the particle swarm algorithm is used to achieve the optimization of the center position and width of the function of the hidden layer of the radial basis neural network, as well as the optimization of the weights between the hidden layer and the output layer, and the optimized model is used for the prediction of bevel cut surface roughness. The optimized model is used to predict the roughness of the bevel cut surface. The experimental results show that compared with the multilayer feed-forward neural network and the standard radial basis neural network model, the model is more accurate in predicting the roughness of the bevel cut, and the coefficient of determination of the prediction model is 0.957 6, the root-mean-square error is 0.032 6, and the average error of deviation is 0.040 9. In this study, we can obtain the prediction model of the roughness of the bevel cut with a high degree of accuracy, and achieve the effective prediction of the roughness of the bevel cut of the high-power laser. This study can obtain a high accuracy prediction model of bevel cutting roughness and achieve the effective prediction of high-power laser bevel cutting roughness.