The neocortex underlies higher cognitive functions in mammals and humans. Its computational abilities are essential for categorizing external objects into supramodal percepts. This natural categorization depends on multisensory integration, to become largely independent of the sensory modality channels through which relevant input is acquired. In contrast, this remarkable emergence of percept invariance is poorly emulated by artificial systems. The anatomical cortical architecture shows extensive connectivity across its different sensory areas, departing from the classically assumed hierarchical processing scheme. Recent studies demonstrated that primary sensory cortical areas coding for distinct modalities are already interconnected, but the computational role of these heteromodal connections is unknown. The goal is to test whether heteromodal interactions in primary sensory cortical areas transmit inferences about the identity of behaviorally relevant objects perceived across multiple sensory channels. The working hypothesis is that these interactions modify the primary representation of unimodal sensory stimuli, resulting in supramodal perceptual invariance, even in ambiguous unimodal contexts. This study will be carried out in awake behaving mice trained to discriminate between two multimodal objects, which require the two sensory modalities to be fully distinguished. State-of-the art techniques (two-photon calcium imaging and multisite electrode arrays) to record from large scale neural assemblies will be combined with modern analysis of neural population dynamics and network simulations. Underlying mechanisms will be explored by optogenetic targeting of specific neuronal populations, to quantitatively challenge mechanistic hypotheses proposed in simulated models. The long-term aim is to quantitatively explain encoding and classification of multisensory cues across primary sensory cortical areas, with the aim of deriving novel generic computational principles by which brain circuits build invariant representations of the environment from ever-changing multisensory input streams.