Inverse effectiveness has been invoked as a principle to describe synergistic effects of multisensory integration in neuronal and behavioural responses as a function of stimulus properties (e.g. intensity) or efficacy. We characterized ‘inverse effectiveness’ and its behavioural relevance at the macroscopic level, as provided by the fMRI BOLD-response, based on (1) stimulus-induced and (2) intrinsic response variability across voxels or subjects during object categorization. Subjects categorized audiovisual object stimuli with the relative informativeness (i.e. degradation) of the auditory and visual inputs being manipulated factorially. Controlling for low-level integration processes, higher-level audiovisual integration was observed selectively in the superior temporal sulci (STS) bilaterally. (1) Consistent with the law of inverse effectiveness, auditory and visual informativeness determined the operational modes of audiovisual integration in STS similarly to the influence of physical stimulus intensity in the superior colliculus: while multisensory interactions were primarily subadditive and even suppressive for intact stimuli, additive effects were observed for degraded, near threshold stimuli. (2) Exploiting intrinsic variability across voxels and/or subjects, we demonstrate that superadditivity for audiovisual stimuli increases with decreasing unimodal responses. This inverse relationship could be explained by inherent statistical dependencies between superadditive and unimodal responses. Nevertheless, the superadditive responses in STS (and only in this region) were related to subjects’ audiovisual behavioral benefit: only subjects that benefited from multisensory integration exhibited superadditive interactions, while those that did not benefit showed suppressive interactions. In conclusion, the (super)additive and subadditive integration modes in STS are functionally relevant and related to behavioral indices of multisensory integration with superadditive interactions mediating successful audiovisual object categorization. We argue that inverse effectiveness trends in neuronal and behavioural responses may be intimately related and mutually predictive.