Magnetic flux leakage (MFL) methods are widely employed for the nondestructive evaluation (NDE) of gas pipelines. The inspection typically generates over 25 GB of compressed data for every 100 mile of pipeline inspected. Currently, trained analysts analyze the data. The gas industry is keenly interested in automating the interpretation process. Key advantages of the automation process include improvement in the accuracy, speed and consistency of interpretation. This paper presents a novel approach to the tasks of analyzing, segmenting and classifying the MFL data from gas pipelines. The analysis is performed in various stages. The first step in the process involves the determination of the size and location of all the MFL indications due to defects and other pipeline artifacts in the data. A list of statistical and physical characteristics of each indication identified in the signal extraction algorithm is compiled. These characteristics, called the signal features, are applied to a hierarchical multi-layer perceptron (MLP) neural network, which classifies the signals. Results from application of the approach to data from field tests are presented. The results suggest that this approach is significantly more effective approach for the classification of MFL data than the conventional back propagation MLP approach.