Organic　Rankine　cycle　(ORC)　can　improve　engine　power　by　recovering　exhaust　energy.　This　paper　co-optimizes　the　engine-ORC　combined　system＇s　power　and　NOx　emission,　with　decision　variables　of　the　engine＇s　excess　air　ratio,　spark　advance　angle,　as　well　as　ORC＇s　pump　and　expander　speeds.　Firstly,　a　simulation　model　of　the　combined　system　is　established　and　validated.　Then,　the　initial　dataset　is　generated　by　the　D-optimum　Latin　hypercube　method　and　simulation　model.　The　artificial　neural　network　(ANN)　prediction　models　of　NOx　emission　and　power　are　established　based　on　these　datasets.　Finally,　the　co-optimization　is　conducted　using　the　ANN　prediction　model　and　genetic　algorithm.　Focusing　on　maximizing　the　combined　system＇s　power　results　in　an　18.30　%　increase　in　power,　and　a　significant　reduction　in　brake-specific　fuel　consumption　(BSFC)　and　brake　specific　NOx　(BSNOx)　by　10.10　%　and　71.30　%,　respectively,　compared　to　the　unoptimized　basis.　Targeting　the　lowest　BSNOx　leads　to　a　limited　1.20　%　increase　in　power　output;　however,　it　results　in　a　19.50　%　increase　in　BSFC.　When　optimizing　for　both　system　output　and　BSNOx,　the　output　remains　13.5　%　above　the　unoptimized　basis.　Meanwhile,　up　to　89.8　%　of　BSNOx　can　be　eliminated　with　negligible　deterioration　in　BSFC.　This　study　could　be　used　for　engine　performance　enhancements.