Defect characterization is a specific procedure of IR thermographic nondestructive testing (NDT) which follows a stage of defect detection. Both procedures can be reference-free or introduce a reference area defined automatically or by the thermographer. The typical reference-free concepts are Pulse Phase Thermography (PPT), Principal Component Analysis (PCA), Thermographic Signal Reconstruction (TSR) and some others. However, by choosing a reference point close to a suspicious (allegedly defect-linked) zone one may consider subtle differences between defect and non-defect areas. Such differences are typically related to differential temperature signals which, in their turn, can be converted into phase shift signals, effusivity variations, etc. In this study, two approaches, namely, analytical formulas and a neural network, have been applied to characterize defects in composites.