Recent evidence from multiple neuroscience techniques indicates that regions within the anterior temporal lobes (ATLs) are a critical node in the neural network for representing conceptual knowledge, yet their function remains elusive. The hub-and-spoke model holds that ATL regions act as a transmodal conceptual hub, distilling the various sensory-motor features of objects and words into integrated, coherent conceptual representations. Single-cell recordings in monkeys suggest that the ATLs are critically involved in visual associative learning; however, investigations of this region in humans have focused on existing knowledge rather than learning. We studied acquisition of new concepts in semantic dementia patients, who have cortical damage centred on the ventrolateral aspects of the ATLs. Patients learned to assign abstract visual stimuli to two categories. The categories conformed to a family resemblance structure in which no individual stimulus features were fully diagnostic; thus the task required participants to form representations that integrate multiple features into a single concept. Patients were unable to do this, instead responding only on the basis of individual features. The study reveals that integrating disparate sources of information into novel coherent concepts is a critical computational function of the ATLs. This explains the central role of this region in conceptual representation and the catastrophic breakdown of concepts in semantic dementia. Copyright © 2013 The Authors. Published by Elsevier Ltd.. All rights reserved.