The brain’s capacity rapidly to integrate information from many different sources lies at the root of our cognitive abilities. Two general questions are particularly relevant to understanding this capacity. First, how is the specialized in- formation conveyed by the activity of functionally segre- gated areas and neuronal groups integrated into a unified, coherent scene? This question underlies one of the oldest controversies in neuroscience – that between localizationist and holistic views of brain function. Second, how is the in- formation that is conveyed by incoming stimuli integrated with information present in memory? It is often assumed that the brain controls behavior by processing incoming stimuli in the form of neural activity patterns. On the other hand, it is also assumed that the brain lays down memories of previously encountered stimuli in the form of patterns of connectivity among neurons. Consequently, information pro- cessing and information storage are often studied separately and with different methodologies. In this review, we consider experimental evidence as well as recent theoretical studies suggesting that the ques- tion of how the brain integrates information can be ad- dressed within a unified conceptual framework. We first discuss known neural mechanisms underlying cognitive and behavioral integration. We then examine a set of statistical measures derived from information theory that can be used to characterize the integration of information among func- tionally segregated groups of neurons. Finally, after consid- ering the key role of spontaneous activity in brain function, we examine how these statistical measures can be used to evaluate the integration of incoming stimuli with ongoing neural interactions. Functional segregation and integration Phrenologists had imagined that different cognitive func- tions were allocated to different parts of the brain well be- fore any neurobiological evidence was available. Advances in neuroscience appear to have fulfilled the phrenologists’ dream and have conclusively demonstrated that functional specialization at multiple spatial scales is a fundamental principle of brain organization. In the visual system, for ex- ample, different brain areas are functionally specialized for visual attributes such as shape, motion, and color 1–3 , and parcellation of function has been discovered within other sensory modalities and in the motor domain 4–6 . Functional segregation extends to the level of columns or groups of neurons 7 . In primary visual cortex neuronal groups are spe- cialized for different stimulus orientations 8,9 , direction of motion 10 and spatial frequency 11 . A similar specialization along different stimulus dimensions has been discovered in essentially every brain area that has been studied in suffi- cient detail 12,13 . Most recently, it has been demonstrated that different brain regions can be activated by specific cog- nitive tasks or by specific stimulus attributes whether these are perceived, imagined, or remembered 14 . While the evidence for regional specialization in the brain is overwhelming, it is clear that the information con- veyed by the activity of specialized groups of neurons must be functionally integrated in order to guide adaptive behavior – just consider how many different signals must be rapidly evaluated and coherently integrated to navigate safely in a busy city. Like functional specialization, functional in- tegration occurs at multiple spatial and temporal scales. In vision, for example, individual elements (dots, edges) are 474 Complexity and coherency: integrating information in the brain Giulio Tononi, Gerald M. Edelman and Olaf Sporns The brains of higher mammals are extraordinary integrative devices. Signals from large numbers of functionally specialized groups of neurons distributed over many brain regions are integrated to generate a coherent, multimodal scene. Signals from the environment are integrated with ongoing, patterned neural activity that provides them with a meaningful context. We review recent advances in neurophysiology and neuroimaging that are beginning to reveal the neural mechanisms of integration. In addition, we discuss concepts and measures derived from information theory that lend a theoretical basis to the notion of complexity as integration of information and suggest new experimental tests of these concepts. G. Tononi, G.M. Edelman and O. Sporns are at The Neurosciences Institute, 10640 John J. Hopkins Drive, San Diego, CA 92121, USA. tel: +1 619 626 2000 fax: +1 619 626 2099 e-mail: tononi@nsi. edu Review Tononi et al. – Complexity and coherency 1364-6613/98/$ – see front matter © 1998 Elsevier Science. All rights reserved. PII: S1364-6613(98)01259-5 Trends in Cognitive Sciences – Vol. 2, No. 12, December 1998