With　integration　of　a　larger　amount　of　clean　power　sources　and　power　electronic　equipment,　operation　and　dynamic　characteristics　of　the　power　grid　are　becoming　more　and　more　complicated　and　stochastic.　Therefore,　it　is　necessary　and　urgent　to　obtain　accurate　real-time　states,　which　is　difficult　from　traditional　state　estimation.　This　paper　systematically　develops　a　phasor　measurement　unit　(PMU)　based　real-time　state　estimator　for　a　realistic　large-scale　power　grid　for　the　first　time.　The　estimator　mainly　relies　on　three　refined　algorithms,　i.e.,　an　improved　linear　state　estimation　algorithm,　a　practical　bad　data　identification　method　and　a　distributed　topology　check　technique.　Furthermore,　a　novel　system　architecture　is　designed　and　implemented　for　the　China　Southern　Power　Grid.　Numerical　simulations　and　extensive　field　operation　results　of　the　state　estimator　recorded　under　both　normal　and　abnormal　situations　are　presented.　All　the　tests　and　field　results　demonstrate　the　advantages　of　the　proposed　algorithms　in　terms　of　online　system　monitoring　and　feasibility　of　refreshing　the　states　of　the　whole　system　at　intervals　of　tens　of　milliseconds.　©　2015　CSEE.