Based　on　the　chaotic　simulated　annealing　network　used　by　Chen　&　Aihara,　which　has　better　searching　properties　than　traditional　Hopfield-Tank　neural　network,　we　proposed　other　two　different　output　functions　and　apply　all　of　them　to　4　city　TSP　respectively.　And　we　constructed　continuous,　discrete　and　continuous-discrete　neural　networks,　introduced　basic　principle　to　initialize　parameter.　At　last,　we　analyze　performance　of　chaotic　neural　networks　with　different　output　function,　which　measured　In　terms　of　feasibility,　research　efficiency,　scalability,　convergence　speed　and　computing　cost.