Pattern recognition is a part of modern information processing with the objective to automate perceptive abilities. There are three paradigms mainly used for practical solutions: neural networks, statistical procedures, and knowledge based systems. Each of them is based on different assumptions and requirements. In our case the automatic recognition and classification of characteristic fringe patterns being indicative for special kinds of flaws on or under the surface of the investigated object is of interest -- especially with respect to an automatic and non-destructive quality control of industrial products. A serious problem of all optical methods is the fact that the used techniques are sensitive only for changes on the surface. Therefore flaws under the surface can be detected just by their affects on the surface. The conclusion from the characteristics in the observed fringe patterns (ring shaped fringes, displacement or distortion of fringes, varying fringe density, ...) to the kind of fault in the measured object (separation, delamination, debond, crack, inclusion, ...) is based on experimental results and practical experience. To solve the task the implementation of two approaches is preferred: neural networks and knowledge based systems. Both approaches have common qualities such as e.g., the preprocessing of noisy interferograms and the selection of representative features but also important differences such as the used recognition architecture. In this paper these aspects are discussed on examples of simple, so-called basic fringe patterns. For these pattern types spot checks are generated using mathematical simulation and practical preparation of loaded samples. Furthermore, the choice of robust features discriminating different basic patterns (classes) and the proposal for a special system architecture are discussed.