Existing energy-saving potential evaluation methods for building occupant behavior improvement do not take occupants' diversity into consideration. This significantly decreases the accuracy of evaluation results. To address this issue, a new energy-saving potential evaluation method has been established. First, principal component analysis is used to reduce the number of influencing factors of occupant behavior. Then, a data mining technique, clustering analysis is used to classify sample occupants into different groups. At last, energy-saving potentials are evaluated in terms of different groups characteristics. This method takes both the impact of various influencing factors and the diversity of occupants into account, and thus could improve the evaluation accuracy significantly. The method was applied to evaluate the energy-saving potential of 100 graduate students' computer standby usage behavior in 3 universities in Hunan province. The results demonstrate its effectiveness and feasibility.