A model predictive control of air temperature in greenhouse based on unsteady CFD model was proposed. Considering the air temperature in greenhouse as the research object and CFD model as a virtual greenhouse environment, the input/output time series data of the system were provided by the unsteady CFD numerical simulation instead of the real physical experiment data. Then the CFD model was converted into system control model based on the data by system identification. The roof ventilation cooling process was controlled by MBPC method based on CFD model. The results showed that the average control error and the standard control error were 2.65℃ and 3.27℃, respectively. The indoor temperature was controlled smoothly and effectively under the allowed range of crop growth. System identification and control algorithm combine with the CFD model improved the efficiency of controller design. This design technique can enrich the technology in greenhouse control system．