Eddy current testing (ECT) is widely used in in-service inspection as well as pre-service inspection of the steam generator (SG) tubes in nuclear power plant of pressurized water reactor type. The interpretation of ECT signals, however, is truly a difficult task so that the reliability enhancement of signal interpretation is strongly desired. An enhanced interpretation tools for ECT signals have been developed by the novel combination of neural networks and finite element modeling for quantitative flaw characterization SG tubes. A database was constructed using synthetic ECT signals generated by the finite element models and principal component analysis (PCA) was adopted in order to optimize the feature set of ECT signals. The improvement in the performances by the features with PCA and the excellent performance for the experimental ECT signals demonstrate the high potential of the developed inversion tools for reliable interpretation of eddy current signals. To explore the possibility of applying the developed approach in practical inspection, we developed an automated system (laboratory prototype) that can acquire experimental ECT signals from SG tubes and carry out the quantitative flaw characterization in a real time fashion by applying the approach developed in the present work.