Accurate prediction of indoor temperature in solar greenhouse is a precondition to accurately control the greenhouse. Because indoor temperature in solar greenhouse is affected by several outdoor environmental factors and the heat conduction mechanism is complex, indoor temperature changes severely in different time. Thus, it is difficult to establish an accurate physical model that describes how outdoor factors affect indoor temperature by mechanism analysis. The accuracy of existing prediction methods based on neural network is low. So，this paper proposed a two-steps method to predict indoor temperature in solar greenhouse based on twice cluster analysis and back propagation (BP) neural network. The first step of the method was two clustering. Similar days were classified to several categories according to clustering of outdoor temperature. Then a whole year training data were split to several continuous time frames.The frames were classified into different categories by clustering of similar days. In the second step, for different categories of time frames, different BP neural networks respectively modeled the relationships between input variables i.e. outdoor temperature, relative humidity, solar radiation, and wind speed and output variable i.e. indoor temperature. The models could be used to predict indoor temperature in solar greenhouse when the outdoor environment was detected. In experiments, two years data was collected from Zhuozhou. For the data, continuous time frames were split to 3 categories.Through the establishment of BP neural network and training respectively, the results show that the prediction error of this method is only 6.23%. Compared with the existing BP neural network prediction algorithm, this method can effectively improve the accuracy, and the average error is reduced by 5.4 percentage points.