We introduce here the Generalized Smoothing Estimator (GSE) for the population mean using nonparametric predicted values based on some auxiliary information in finite population survey sampling. This estimator is in the spirit of the Generalized Regression Estimator (GRE) due to Cassel, Sarndal and Wretman (1976, Biornetrika). Where as the GRE uses model based prediction. GSE uses the predicted values of the non-parametric regression technique, such as the kernel or spline methods. Some asymptotic properties of the estimator are studied along with an empirical study for the comparison of GSE and GRE. It is found that GSE might produce large gains in efficiency in the face of serious model mis-specification.