Abstract: To systematically analyze the relationship of climate characteristics inside and outside the solar greenhouse, and provide support to the solar greenhouse crop environmental regulation and microclimate forecasting, based on the observed data of plastic sunlight greenhouse microclimate and neighboring weather station, by using the method of BP neural network, 3 models were established, to assimilate the microclimate characters under plastic sunlight greenhouse in Shijiazhuang Region during the winter. The results showed that: all of the root mean square error (RMSE) between air temperature trained and measured value from 3 models was no more than 2℃ and the coefficient of determination was more than 0.95 respectively. RMSE between relative humidity trained and measured value was no more than 2 percent points and the coefficient of determination was more than 0.95. RMSE between trained and measured value of solar radiation received was no more than 16 W/m2 and the coefficient of determination was also more than 0.95. All of the RMSE between air temperature predicted and measured value from the 3 models was no more than 2℃ and the coefficient of determination was more than 0.95 respectively. RMSE between relative humidity predicated and measured value was no more than 4 percent points, and their coefficient of determination was more than 0.9 in sunny or slight cloud-cloudy day, more than that which was 0.8 approximately in sunless day. RMSE between predicted and measured value of solar radiation received was no more than 26 W/m2 and the coefficient of determination was more than 0.95. The results indicated that 3 BP neural network models had quite precisely for predicting microclimate characteristics under plastic sunlight greenhouse in different weather conditions in Shijiazhuang Region in winter, which could meet the forecast requirements.