In this paper, the regression prediction models of surface reference temperature and softening layer depth are obtained through preheating simulation and experimental verification based on the surface response methodology (RSM), and the simulation predictions of surface reference temperature and softening layer depth are optimized. Firstly, a three-dimensional transient heat transfer model is established for turning SiC ceramics as the research object, and the concept of surface reference temperature is proposed. The single factor simulation and experiment are carried out by changing the laser power, and the simulation value of surface reference temperature and the experimental value are compared to verify the effectiveness of the heat transfer model. Then, the material removal mechanism is explained according to the material fracture theory, and the concept of softening layer depth is proposed. The multivariate prediction models are established to obtain the surface reference temperature and the softening layer depth based on the RSM by selecting three factors: laser power, spot diameter and rotational speed. Finally, the desirability function analysis (DFA) is used to compare the measured value of surface reference temperature with the predicted value, which proves that the regression prediction model is effective. The range of laser power for single-factor experiment of surface reference temperature is 175 W–235 W, and the error of the results is between 7.57% and 10.82%. The Box-Behnken experimental design is adopted with the following factors and levels: laser power (200 W–230 W), spot diameter (1 mm–1.4 mm) and rotational speed (1080r/miñ2160r/min); the regression models are analyzed by residual analysis, ANOVA, 3D surfaces, and contour plots to prove the reliability. Based on the DFA method to select the level of each factor: laser power of 205 W and 220 W, spot diameter of 1 mm, and rotational speed of 1620r/min, the errors between the optimal and experimental values of surface reference temperature are 9.82% and 7.72% respectively, which not only achieve the optimization of the desirability function of the heat transfer model, but also prove that the regression prediction model has good agreement with the experimental values. The regression prediction model of softening layer depth establishes the functional relationship between softening layer depth and different factors in the study range and provides direct reference for the selection of cutting depth in laser assisted machining (LAM) SiC ceramics.