We　propose　a　Spatial　Artificial　Neural　Network　(SANN)　with　spatial　architecture　which　consists　of　a　multilayer　feedforward　neural　network　with　hidden　units　adopt　recurrent　lateral　inhibition　connection,　all　input　and　hidden　neurons　have　synapses　connections　with　the　output　neurons.　In　addition,　a　supervised　learning　algorithm　based　on　error　back　propagation　is　developed.　The　proposed　network　has　shown　a　superior　generalization　capability　in　simulations　with　pattern　recognition　and　non-linear　function　approximation　problems.　And,　the　experimental　also　shown　that　SANN　has　the　capability　of　avoiding　local　minima　problem.　©　2011　Springer-Verlag.