Nowadays, in many branches of mass production, it is often important to obtain the highest quality surfaces. Hence, it is necessary to have reliable and precisions methods of surface quality control. For this purpose, interferometric methods are widely used. This study is devoted to adequate modeling of monochrome interference patterns of optical surfaces with typical defects to solve the problem of automatic surface quality control. The created parametric models of the Linnik microinterferometer and the surface defects were used to generate a synthetic dataset for training and testing an artificial neural network. The constructed and trained neural network allowed the classification of five typical defects classes with 85% accuracy for synthetic patterns with distortions and 72% accuracy for real images of interference patterns.