Tomography is an important technique for noninvasive imaging: images of the interior of an object are computed from several scanned projections of the object, covering a range of angles. Traditionally, tomographic reconstruction algorithms have been based on techniques from analysis and linear algebra. In this paper we describe how a particular version of the tomographic reconstruction problem, known as discrete tomography, can be considered as a classification problem. By making this connection between two seemingly unrelated scientific areas, the full machinery of learning classifier theory can be used to solve tomographic problems. The use of classifiers that can be trained from examples for tomography has two main advantages. First, prior knowledge concerning the images of interest can be learned automatically. Second, there are several types of classifiers that can perform the classification task very fast once trained. Real-time reconstruction is one of the main goals in tomograp...