The paper deals with estimating problem of regression function ata given state point in nonparametric regression models with gaus-sian noises and with nongaussian noises having unknown distribution.An asymptotically efficient kernel estimator is constructed for alocal minimax risk.