We present results on a priori estimation of LES wall models by nonlinear stochastic estimation (NLSE). The results are given in form of nonparametric, generalized additive models. First results on a posteriori LES are shown. The analyzed data are taken from a DNS of plane channel flow. The applied method allows for quantitative identification of important model terms and reveals physical mechanisms of the near-wall region. Investigated models are generalizations of known algebraic models. With usual linear methods the resulting functions cannot be found. Additional terms like the wall-normal velocity component and the streamwise pressure gradient are included in the wall model. We find that different terms become important in different regions near the wall.