Abstract: In order to realize the non-destructive identification of wall paints, a method of fast and non-destructive identification of wall paints by microscopic confocal Raman spectroscopy and multiple modeling was proposed. The influence of Savitzky-Golay(SG) smoothing polynomial and points, and compared the identification ability of different models were investigated. The results showed that compared with radial basis function neural network model, multilayer perceptron neural network model has a stronger ability to identify samples. Different brands of wall paints have been identified exactly by multilayer perceptron neural network model after SG smoothing 1-degree polynomial and smoothing points of 27 points. At the same time, primers, surface coatings and varnishes of Meffert samples were also identified accurately. This method improved the efficiency of identification, reduced the cost, which is worth consulting.