According to an experimental dataset on the softening points of 30 bitumen samples under different resistances and temperatures, the support vector regression (SVR) approach combined with particle swarm optimization (PSO) for its parameter optimization is proposed to conduct leave-one-out cross validation (LOOCV) for modeling and predicting the softening point of bitumen, and its prediction result is compared with that of multivariate linear regression (MLR). The maximum error 2.1 ℃ predicted by SVR is much less than 7.9 ℃ which is calculated by MLR modeling. The statistical results reveal that the root mean square error (RMSE=0.75 ℃), mean absolute error (MAE=0.32 ℃) and mean absolute percentage error (MAPE=0.28%) achieved by SVR-LOOCV are all less than those (RMSE=3.3 ℃,MAE=2.6 ℃ and MAPE=2.34%) calculated via MLR model. This study suggests that the softening point of bitumen can be forecasted timely by SVR to provide an accurate guidance for producing of high-quality bitumen.