This paper studies the performance prediction of ground source heat pump (GSHP) systems by real-time monitoring data and data-driven models. A GSHP system, which is installed in an office building of Shaoxing (29.42°N, 120.16°E), China, is real-time monitored from Nov. 2012 to Mar. 2015. Data mining (DM) technologies were simultaneously applied to process the monitoring data and find the required inputs for data-driven models. Back-propagation Neural Network (BPNN) algorithm was selected from six classical sorting algorithms to establish the data-driven models. The performance of the GSHP system from Nov. 2012 to Mar. 2015 was evaluated by the monitoring data. And the long-term performance was predicted by the data-driven models. The monitoring results show that the application effectiveness of the GSHP system is unsatisfied because of the high pumping power. Moreover, the relationship between the short-term and long-term performance of GSHP system is investigated for the purpose of predicting the long-term performance of GSHP system by a short-term monitoring data. The monitoring data of different days in several modes are needed to predict the long-term performance of GSHP system under a certain deviation.