Building　energy　efficiency　is　an　important　section　of　energy　conservation,　while　the　premise　of　mining　energy-saving　potential　is　clearly　obtain　the　specific　energy　consumption　of　each　building　service　system.　So　electricity　sub-metering　was　widely　implemented　in　China　in　recent　years.　Main　building　energy　systems　and　high-power　electrical　equipment　all　can　be　metered　directly　except　the　HVAC　terminal　units　(like　fan　coils　and　air　handing　units),　because　they　are　always　mixed　with　lighting　system　caused　by　existing　electricity　circuit　design.　It　is　an　obstacle　to　get　detailed　and　systematic　sub-metering　data.　Therefore,　an　indirect　calculation　algorithm　is　necessary.　Considering　the　fact　that　the　amount　of　historical　sub-metering　data　is　huge,　the　algorithm　should　be　calculated　fast　and　ensure　a　certain　degree　of　accuracy.　And　given　the　variety　of　building　types,　it　is　better　that　the　algorithm　also　has　classification　function.　To　solve　above　problem,　an　automatic　HVAC　terminals　energy　use　calculation　method　is　proposed　in　this　paper.　The　core　algorithm　of　this　method　is　classification　and　regression　tree　(CART).　In　this　research,　the　establishment　of　CART　model　is　achieved　and　the　selection　of　input　variables　is　discussed.　Finally,　this　method　is　demonstrated　and　validated　in　an　actual　building　located　in　Shanghai.