In the long-term forecasting work, the original data has the characteristics of randomicity and non-linear movement, and also the capacity of available study samples is small and information is insufficient. The bayesian-regularization neural network possesses the characteristics of strong nonlinear fitting and the capabilities of excellent generalization. Unbiased GM(1,1) can use few data to construct models, it can weaken the randomicity of the original data and strengthen regularity, and also can eliminate the inherent deviation of the conventional GM(1,1) model. Making the best use of the merits of the two, the combined model of unbiased GM(1,1) and bayesian-regularization neural network are constructed and put into real traffic forecasting work. By contrasting with BP network, the result shows that this model is feasible and efficient, the accuracy of forecasting is also increased.