Reinforcement learning refers to improving performance through trial-and-error. Despite recent progress in developing artificial learning systems, including new learning methods for artificial neural networks, most of these systems learn under the tutelage of a knowledgeable 'teacher' able to tell them how to respond to a set of training stimuli. Learning under these conditions is not adequate, however, when it is costly, or even impossible, to obtain this kind of training information. Reinforcement learning is attracting increasing attention in computer science and engineering because it can be used by autonomous systems to learn from their experiences instead of from knowledgeable teachers, and it is attracting attention in computational neuroscience because it is consonant with biological principles. Recent research has improved the efficiency of reinforcement learning and has provided some striking examples of its capabilities.