In　this　paper　empirical　likelihood　(EL)-based　inference　for　a　semiparametric　varying-coefficient　spatial　autoregressive　model　is　investigated.　The　maximum　EL　estimators　for　the　parametric　component　and　the　nonparametric　component　are　established.　Furthermore,　asymptotic　properties　of　the　proposed　estimators　and　EL　ratios　are　derived,　and　the　corresponding　confidence　regions/bands　are　constructed.　Their　finite　sample　performances　are　studied　via　simulation　and　an　example.