With　great　development　in　numerical　optimization　and　wide　application　of　computational　fluid　dynamic　methods,　the　three-dimensional　design　optimization　of　multi-stage　axial　flow　compressor　will　become　a　routine　practice　in　the　future.　The　common　critical　issue　arisen　from　the　practice　is　how　to　carry　out　the　three-dimensional　design　optimization　of　multi-stage　axial　flow　compressor　at　an　affordable　design　cost.　To　tackle　this　problem,　a　three-dimensional　design　optimization　method　is　developed,　in　which　reference　vector　guided　evolutionary　algorithm,　the　ensemble　of　surrogate　models,　adaptive　sampling　method　and　computational　fluid　dynamics　(CFD)　simulation　are　integrated,　to　improve　the　adiabatic　efficiency　and　broaden　the　stability　margin.　The　ensemble　of　surrogate　model　consists　of　linear　regression　model,　support　vector　regression　model,　Kriging　model　and　radial　basis　function.　The　blade　lean　and　twist　are　set　as　design　variables　for　all　the　stators　while　for　the　rotors　only　the　blade　twist　is　considered.　By　taking　advantage　of　these　advanced　methods,　a　6.5-stage　axial　flow　compressor　with　totally　60　design　variables　is　optimized　at　design　speed-line.　CFD　calculations　show　that　the　adiabatic　efficiency　at　design　point　is　increased　by　0.88%.　The　stall　margin　is　increased　from　12.72%　to　14.08%,　and　the　choke　margin　is　increased　from　20.41%　to　21.35%.　Flow　analysis　of　the　optimal　compressor　indicates　that　the　blade　lean　is　mainly　responsible　for　suppressing　the　endwall　flow　separation　while　the　blade　twist　for　the　stage　match　between　adjacent　blade　rows.　This　work　is　of　great　significance　for　three-dimensional　multi-objective　design　optimization　of　multi-stage　axial　flow　compressor.　Copyright　©　2022　by　ASME.