Gas transmission pipelines are routinely inspected using a magnetizer-sensor assemblage, called a pig, which employs magnetic flux leakage (MFL) principles to generate defect signals that can be used for characterizing defects in the pipeline[1]. Previously reported work[2] demonstrated that radial basis function(RBF) networks[3–5] can be employed to characterize MFL signals in terms of defect geometry. Further development of this research work, related to three dimensional defect characterization are reported elsewhere in these proceedings. This paper presents an alternate neural network approach based on wavelet functions to predict three dimensional defect profiles from MFL indications. Wavelet basis function neural networks are comprised of a hierarchical architecture and are capable of multiresolution functional approximation. They offer a powerful alternative to RBF based signal-defect mapping techniques, in that the level of output prediction accuracy can be controlled by the number of resolutions in the network architecture. Consequently, the network itself can be employed to generate measures of confidence for its prediction. Such confidence factors may prove to be extremely useful in pipeline inspection procedures since they can form a basis for subsequent remedial measures. The feasibility of employing a wavelet basis function network for characterizing defects in pipelines is demonstrated by predicting defect profiles from experimental magnetic flux leakage signals.