The accuracy and stability of the short-term electricity price forecasting is very important for electricity bidding decisions. Aimed at disadvantages of the normal short-term electricity price forecasting methods， an intelligent forecasting method based on multi-agent is proposed， in which the intelligent model selection as well as the combination of quantitative and qualitative forecasting results given by different models are implemented based on the idea of intelligent forecast and combining forecast. The construction of knowledge base and the forecasting flow based on multi-agent are emphasized. Simulation results show that the intelligent model provides better accuracy and stability compared with normal models.