Under some conditions, it is proved t hat the kernal estimators converges to the regression function at the rate of th e law of the iterated logarithm.