In order to improve the quality and reliability of current maintenance decision processes used on carbon fibre-reinforced polymer (CFRP) structures, automated evaluation procedures come into play. The non-destructive testing (NDT) technique considered in this paper is active thermography with long pulse excitation. The objective of this paper is to present a multisource delamination detection and depth estimation system based on learning machines for CFRP structures. The proposed technique pre-processes the thermographic response to produce three different information sources: the normalised thermographic response, the normalised temporal contrast and the phase contrast. Out of these information sources several features are extracted, which are subsequently used for classification. In this paper, two different classifiers are considered: a binary classifier for defect detection and a multiclass classifier for depth estimation, both based on support vector machines (SVMs). The system is trained using prepared CFRP samples and tested using a different set of CFRP samples with impact damage. Very good detection and depth estimation results are achieved for impact delaminations down to a 2 mm depth. [ABSTRACT FROM AUTHOR]