Cross-modal illusions such as the McGurk-MacDonald effect have been used to illustrate the automatic, encapsulated nature of multisensory integration. This characterization is based in the widespread assumption that the illusory percept arising from intersensory conflict reflects only the end-product of the multisensory integration process, with the mismatch between the original unisensory events remaining largely hidden from awareness. Here the authors show that when presented with desynchronized audiovisual speech syllables, observers are often able to detect the temporal mismatch while experiencing the McGurk-MacDonald illusion. Thus, contrary to previous assumptions, it seems possible to gain access to information about the individual sensory components of a multisensory (integrated) percept. On the basis of this and similar findings, the authors argue that multisensory integration is a multifaceted process during which different attributes of the (multisensory) object might be bound by different mechanisms and possibly at different times. This proposal contrasts with classic conceptions of multisensory integration as a homogeneous process whereby all the attributes of a multisensory event are treated in a unified manner. (Contains 6 footnotes and 1 figure.)