A method has been developed for the classification of underwater target geometry using bistatic acoustic amplitude data collected by an Autonomous Underwater Vehicle (AUV) as it follows a selected path through the scattered field created by a fixed source insonifying a target. The mobility of an AUV allows it to exploit features of this field in three dimensions. The classification method presented includes offline and onboard processing components, which use a combination of signal processing, vehicle behaviors, and machine learning in the form of Support Vector Machines (SVMs) to extract target geometry from collected acoustic data. The offline training and analysis step creates training and test vector sets in a selected feature space from existing scattered field data and outputs models for target classification, confidence, and feature ranking. Several algorithms are explored for selecting the feature space used by the SVM. The models produced by the offline processing step are used in the real-time classification processing chain onboard an AUV sampling an unclassified target's scattered field. The presented simulation results use scattered fields modeled using OASES-SCATT and demonstrate real-time processing and path planning in the LAMSS MOOS-IvP simulation environment. [Work supported by ONR Code 321 OA and NSF GRFP.]