Abstract: This study describes a ceramic wall tiles surface quality control classification training algorithms. The algorithm employed statistical approach based on Bayes decision functions and minimum distance techniques for classification. The measured feature vectors of the training tile samples are used by the algorithm to generate 2-D display of the classifier in feature space. Euclidean distance between each of the training tile samples and the test tile sample in the feature space is computed by the algorithm. The test sample is assigned the class to which it is closet. Many experiments were conducted using different number of defective test tile samples ranging from 50-150 samples. The classification of these defective samples for three sets gave an average of 1.45% error rate. This classification performance is better than human operator within the shorted time taken by the machine.