Abstract: Two five-star hotels in Sanya were selected to be studied.The main load factors of hotels were found out according to the measured data,and chilled water temperature was an important load factor which should be considered in load predition.Applying general regression neural network methodology,this paper set up a multi-outputs load prediction model,and input vector was designed to be five cases.Two real data sets from two hotels were used to test the load model.The result demonstrates that chilled water temperature plays an important role in the load prediction,and its use can significantly improve the predictive accuracy.The GRNN model,which using the last 24 hours load data,the next day weather forecastings and chilled water temperatures as the input,and the next 24 hours loads as the output,is found to be effective for load prediction,and it is simple to make in practical utilization.