How　to　control　the　growth　of　building　energy　consumption　and　achieve　the　goal　of　energy　saving　and　emission　reduction　while　ensuring　people＇s　growing　demand　for　indoor　comfort　is　of　great　practical　significance　in　the　new　era.　The　rapid　and　accurate　prediction　of　the　building　energy　consumption　at　the　early　design　stage　can　provide　a　quantitative　basis　for　the　energy-saving　design.　ANN　(artificial　neural　network)　model　is　the　most　widely　used　artificial　intelligence　model　in　the　field　of　building　performance　optimization　due　to　its　high　speed,　high　accuracy,　and　capability　of　handling　nonlinear　relationships　between　variables.　In　this　paper,　an　ANN-based　fast　building　energy　consumption　prediction　method　for　complex　architectural　form　for　the　early　design　stage　was　proposed.　Under　this　method,　the　authors　proposed　an　idea　of　architectural　form　decomposition,　to　eliminate　the　complexity　of　building　shape　at　the　early　design　stage,　thus　transforming　the　energy　consumption　prediction　problem　of　one　complex　architectural　form　into　several　energy　consumption　prediction　problems　of　multiple　simple　blocks:　the　method　of　characterization　decomposition　(MCD)　and　the　method　of　spatial　homogenization　decomposition　(MSHD).　The　ANN　model　was　introduced　to　realize　energy　consumption　prediction,　which　fully　utilized　the　two　advantages:　high　speed　and　good　response　to　complicated　relationships.　Accuracy　verification　shows　that　the　relative　deviation　of　cooling　and　heating　energy　consumption　is　within　+/-　10%　using　the　MCD　method.　The　relative　deviation　of　total　energy　consumption　is　within　10%　using　the　MSHD　method.