In this paper a neural network based approach, applicable in real time applications, is proposed for detecting the number and location of defective elements in a typical uniformly excited microstrip planar array antenna fed in a serial manner. Here the defective elements are those elements that are not excited by the feed lines but radiations due to the induced currents in the surface of these elements still remain. The neural network performs a nonlinear mapping between some samples of the degraded patterns of the simulated array antenna and the array elements which may have caused these degradations. After the training procedure the proposed fault diagnosis system is very fast and has a satisfactory success rate.