Context plays a critical role in the way the human brain processes information. Our perception of sensory inputs depends strongly on the spatial and temporal history of the recent sensory past. Context sensitive sensory responses can be understood in terms of adaptive gain control mechanisms, where estimates of variability (e.g. uncertainty or confidence) scale the driving neural responses to sensory input via feedback connections and the action of neuromodulators. Such computations are at the heart of predictive coding accounts of brain function, and offer a formal neurobiological framework to interrogate the long standing notion of ‘context insensitive’ perception in autism (Lawson et al., PNAS, 2015). In this talk I will first introduce the framework by which estimated variance, or precision-weighting, may be aberrant in autism (Lawson et al, Front Hum Neuro, 2012). Then I will go on to present some recent data examining how individuals with autism learn about the variability of different kinds of sensory information and how these learning dynamics are related to behavioural responses, symptom severity and neuromodulatory function.