For the accuracy of indoor temperature prediction method for large cooling building is not high enough to fulfill the requirement of energy optimal control for HVAC system, an equivalent thermal model of building and a recursive least squares - Kalman filtering method (RLS-KF) for indoor temperature prediction are proposed in this paper. In order to describe the unsteady state thermal characteristics of the building, a three order building thermal model is established by equivalent circuit method, and air-conditioning cooling load, ambient temperature and solar radiation intensity are selected as input variables of the model. The RLS method is used to identify the model parameters online, however, aiming at the low prediction accuracy of single RLS method, a pseudo-measurement value is constructed, and the KF algorithm is applied to the room temperature prediction problem. Taking an office building in Guangdong as an example, the results show that the prediction accuracy and stability of RLS-KF algorithm is much higher than that of a single RLS method, and the performance is better at short-term room temperature prediction.