Constant-speed　air-cooled　water　chillers　(CAWCs)　are　typically　on-off　controlled　in　accordance　with　a　fixed　set　point　of　chilled　water　return　temperature　(T-wr),　which　seriously　restricts　their　efficiency　during　part　load　conditions.　To　improve　the　performance　of　CAWC　air　conditioning　(A/C)　systems,　this　paper　developed　an　optimal　control　method　to　reset　the　T-wr　of　CAWCs　real　time.　Firstly,　a　simulation　platform　was　established　based　on　an　actual　office　building　with　a　CAWC　A/C　system　using　fan　coil　units　(FCUs)　as　terminal　devices　in　Beijing.　Secondly,　a　dataset　of　optimal　values　of　at　different　part　load　conditions　was　calculated　by　simulation.　Thirdly,　to　enable　the　practical　use　of　the　dataset,　a　general　regression　neural　network　(GRNN)　model　was　built　to　predict　the　optimal　T-wr　value　according　to　three　ambient　parameters　and　one　operation　parameter.　Finally,　a　GRNN　based　optimal　control　method　was　developed　for　the　CAWC/FCU　A/C　system.　The　numerical　results　demonstrated　that　the　proposed　method　can　achieve　a　good　control　over　indoor　thermal　environment,　and　lead　to　a　reduced　energy　use　for　the　A/C　system　by　11.0%　over　the　cooling　season.　This　study　provided　a　promising　control　method　of　chilled　water　temperature　for　CAWC/FCU　A/C　systems　in　dry　climates.