This work presents an approach to extract, analyze and select image attributes necessary for building a vision computer system that is able to detect and classify tufted carper defects. Four different tufted carpets and four different defects were considered; missing pile, higher pile, lower pile, slobs and knots. Carpet digital images were statistically analyzed to calculate Mean, Variance, Skewness, Kurtosis, and Entropy, known as tonal features. Texture features were extracted from co-occurrence matrices describing the relationships between intensities of two pixels at a certain distance and angle from each other and evaluated using SGLDM and GLDM statistics. Graphical presentation and visual assessment were made to choose the most significant features. For classification. artificial neural networks were built and trained using perceptron and back propagation algorithms. The recognition was successful in detecting common tufted carpet defects.