The question "What makes a brain autistic?" admits at least two senses: one diagnostic, another ætiological. Answering the ætiological question demands that we avoid mistaking the developmental endpoints that are diagnostic of autism for the ætiological roots that are causative, and that we frame our hypotheses and experiments with this distinction in mind. Much like the people whom it seeks to understand, autism research is all too prone to focus on surface details and to ignore the complexities of the overall picture: in brain anatomy and physiology, the success of lesion models in understanding disorders of the adult brain has led to a focus on cognitive and anatomical modules which neglects developmental interactions, and in genetics successes in single-gene disorders have produced a similar neglect of gene-gene interactions. The very complexity and heterogeneity of findings suggest that autism's final common pathway will be found not in single genes nor in single brain regions, but rather at the more abstract and pervasive level of neural information processing. Such an abnormality of neural networks could arise from convergence of multiple and various genetic and environmental factors, and would produce a divergence into heterogeneous symptoms, severities, and endophenotypes. A confluence of neuroanatomical, neuropathological, genetic, and neurophysiological results motivates the hypothesis that this network abnormality may consist of a skewed balance between local and long-range neural connectivities, in which excessive local connectivity decreases information capacity, and sabotages normal activity-dependent development of long-range interactions. Anatomically, the autistic brain grows abnormally large early in postnatal development, and short-range white-matter tracts are particularly hypertrophied – a phenomenon likely related to neuropathological results of altered minicolumn structure and to linkage with genes affecting neural connectivity. Physiologically, our own and others' results indicate an absence of selective, top-down modulation of neural subsystems, and development of compensatory strategies that rely on abnormally intense processing of low-level sensory and perceptual representations. This functional pattern is in general associated with abnormally low activation of prefrontal and other integrative brain regions, abnormally high activation of unimodal regions, and abnormally low functional connectivity between regions. Interestingly, many non-autistic sibs of children with autism manifest the frontal deactivation but not the unimodal hyperactivation. This physiological finding complements the large corpus of behavioural results on the Broader Autism Phenotype in first-degree relatives – a population that can be informative precisely because ætiological factors are not obscured by all the developmental sequelae of the full syndrome of autism.