With　an　increasing　number　of　Geographical　Information　System　(GIS)　services　publicly　available　on　the　Web,　the　discovery　of　composite　GIS　services　is　promising　when　novel　requirements　are　to　be　satisfied.　GIS　services　in　the　repository　like　ArcGIS　software　are　organized　in　a　tree　hierarchy,　where　a　parent　node　represents　a　categorial　GIS　service　with　a　coarser-granularity　than　its　child　GIS　services,　while　leaf　nodes　correspond　to　atomic　and　exercisable　GIS　services.　In　this　setting,　discovering　appropriate　atomic　GIS　services　is　challenging.　To　remedy　this　issue,　this　paper　proposes　a　composite　GIS　service　discovery　mechanism.　Specifically,　for　the　given　requirement,　select　the　parent　nodes　that　take　the　given　input　parameters　as　input　and　remove　their　inactivated　children.　Use　remaining　children　to　build　the　network　and　repeat　the　previous　operation　until　finding　the　services　that　contain　the　required　output.　Then　record　the　semantic　similarity　degree,　calculated　by　services　functional　description,　in　this　network.　By　using　the　simulated　annealing　algorithm,　a　composite　GIS　services　solution　will　be　recommended　from　this　semantic　network.　Evaluation　results　demonstrate　that　our　approach　could　give　more　significant　solution　compared　with　the　state-of-the-art　techniques.