A　neural　sliding　mode　controller　is　presented　for　trajectory　tracking　control　of　multi-link　robots　with　uncertain　external　disturbances　and　system　model　errors.　This　approach　gives　a　new　global　sliding　mode　manifold　for　the　second-order　multi-link　robots,　which　enable　system　trajectory　to　run　on　the　sliding　mode　manifold　at　the　initial　states　and　eliminate　the　reaching　phase　of　conventional　sliding　mode　control.　Robustness　for　system　dynamics　is　guaranteed　over　all　the　response　time.　A　radial　basis　function　neural　network　(RBFNN)　is　employed　to　eliminate　chattering　of　global　sliding　mode　control,　and　enforce　the　sliding　mode　motion　by　its　learning　the　upper　bound　of　model　errors　and　uncertain　disturbances.　Moreover,　the　stability　of　the　controller　is　proven　by　Lyapunov　function.　Simulation　results　verify　the　validity　of　the　control　scheme.