Aiming at the problem of poor efficiency of ceramic tile production caused by the mismatch between higher and higher speed of production and slow speed of artificial classification, the paper presented an algorithm about extracting the features of color and texture of ceramic tiles and an algorithm about improved multilayer perceptron neural network(MLPNN) aiming at the problem of multi-classification based on machine vision software, HALCON 11.0, as the development platform. Firstly, the images of ceramic tiles were denoised as pretreatment. Then the system extracted the hue features of ceramic tiles in HSI color space, calculated the gray level co-occurrence matrix(GLCM) and gray level characteristics of amplitude distribution to reflect the texture feature of ceramic tiles, and put the features as input layer neurons of multilayer perceptron neural network. Next, the paper designed the multilayer perceptron neural network with putting softmax function as the activation for pattern matching, and compared with the pattern matching method of BP neural network. Finally, an experimental prototype of classification system was built with simple user interface. The experimental results show that, the classification accuracy for all kinds of ceramic tiles in the experiments are over 90%. The system has high classification accuracy for the random texture ceramic tiles and can be applied to production of ceramic tiles.