Abstract: Based on the meteorological data both inside and outside the plastic greenhouse in Cixi, Zhejiang province and agricultural meteorological experimental station of Nanjing University of Information Science and Technology, three BP neural network models were established, which the input variable was chosen as radiation solar outdoor, air temperature, relative humidity and wind speed, and output variable was chosen as temperature indoor and relative humidity. The results showed that all of the root mean square error (RMSE) between trained air temperature and measured value from three models was no more than 2℃ and the relative error (RE) no more than 4% respectively. Both RMSE and RE between trained relative humidity and measured value was no more than 7 percent points and 7%. All of the RMSE between predicted air temperature and measured value from three models was 2℃ approximately, and their RE was no more than 6% in spring, less than that in winter. RMSE and RE predicated relative humidity and measured value was no more than 7percent points and 9% respectively. The results indicated that three BP neural network models had quite precisely for predicting temperature indoor and relative humidity in plastic greenhouse, which could meet the forecast requirements for plastic greenhouse microclimate.