Let (Xj, Yj),j = 1,n, be a sample from the populatin (X, Y) with an unknown joint probability density f(x,y) and g(x) be the marginal density of X, which is supposed to be known. The recursive formula τn(x) = n-1ΣjajYjK(aj(x-Xj))g-1(x) is considered as an estimator of the regression function τ(x) = E(Y/X = x) and asymptotic distribution of the deviation nan-1 ∫(τn(x)- τ(x))2g2(x)h(x)dx is investigated. © 1990.