Forensic evidence comes in many forms, but almost always measurements are high-dimensional. As an example, in an image of a shoe out-sole, the data consist of the coordinates of between 8,000 and 10,000 edge pixels depending on pattern and size. Forensic scientists are typically interested in questions of source: do two items have the same (perhaps unknown) source? From a statistical viewpoint, this is a classification problem that can be effectively addressed using learning algorithms. We apply learning algorithms to compute the likelihood that a specific shoe may have left the shoe print found at a crime scene.