The purpose of this study is to study contract capacity of power demand by utilizing gray prediction theory. Historical data for yearly power demand for a high-voltage user in the Hsinchu area is collected and analyzed. Optimal annual contract capacities from 2005 to 2011 are derived. Furthermore, prediction, analysis and comparison of the derived data are carried out for four different methods, namely the rolling forecast method, moving average method, MegaStat method and gray prediction method. The results show that, under certain conditions with limited data, the power demand prediction utilizing gray prediction is most accurate, compared to the other three methods. To compare actual data and prediction results, average absolute deviation percentage is used as the prediction accuracy criterion. It is found that the gray prediction method performs well in Taiwan power demand prediction. Therefore, the contract capacity prediction for the year of 2012 utilizing gray prediction method can be used as a reference for future electricity budget planning.