This　work　investigates　the　state　estimation　problem　in　a　network　of　diffusively　coupled　neural　networks　(NNs),　where　each　NN　has　delayed　dynamics,　and　the　designer　can　only　measure　the　states　of　a　portion　of　NNs.　In　light　of　the　master-slave　synchronization　framework,　an　estimator　network　of　NNs　incorporating　pinning　control　is　constructed,　and　the　control　gain　is　assumed　to　be　changeable.　Some　sufficient　criteria　are　established　for　successful　state　estimation,　which　reveal　the　influence　of　NNs＇　dynamics,　network　topology,　and　time　delays.　Moreover,　some　corollaries　are　also　derived　to　cater　several　specified　cases.　Besides,　a　criterion　independent　on　the　exact　topological　information　is　also　given　to　tackle　scenarios　with　topological　uncertainties.　Finally,　some　numerical　examples　are　provided　to　demonstrate　the　validity　of　theoretical　results　and　the　feasibility　of　control　schemes.