Results of new studies of C-scan imaging on austenitic welds, used in conjunction with artificial neural network techniques, are reported. The C-scan technique provides high-resolution images of subsurface zones which are inaccessible by conventional A-scans. Based on the anisotrophy-induced perturbation of ultrasonic signals, the anisotropic regions in various steels were detected by C-scan imaging. C-scan images of both homogeneous samples of austenitic steels and heterogeneous samples of austenitic welds allowed the neural network-based ‘‘training‘‘ of the measurement and analysis system. The locations and approximate compositions of the various inhomogeneities present in the welds were determined. Image analysis was used to identify three austenitic structures (W.4541, W.6903 and HP 50) in the welds under study.