To realize friendly interaction between building heating, ventilation and air conditioning（HVAC）and the power grid, a demand response cost assessment method combining model and data driven is proposed. Firstly, the fitting ability of artificial neural network（ANN） is used to automatically learn from the physical model and build a dynamic model of HVAC. Then, considering the impact of temperature changes on the personnel efficiency of building, an optimization model of the HVAC setting temperature is established and solved based on particle swarm optimization（PSO）algorithm. Finally, additional cost incurred by users due to deviation from the optimal set temperature is defined as response cost, and demand response cost assessment for HVAC is proposed accordingly. Simulation results show that by setting optimal temperature of HVAC, total cost of users can be significantly reduced. The response cost curve can reflect users’productivity loss factors, thereby helping users determine the bid price rationally in demand response.