Modality switching costs (MSCs) are one of the classic effects that support the embodied views of conceptual representations. They refer to a delay in response time to verify a sensory property of a certain perceptual modality (e.g., visual: SUN – bright), when the previous sensory property has been of a different modality (e.g., auditory: BLENDER – loud) compared to a property of the same modality (e.g., visual: ROSE – red). Such costs indicate that conceptual representations require the recruitment of modality-specific resources. However, MSCs could also result from the distributional pattern of property words: the reason why loud -> bright takes longer than red -> bright could be because bright and loud do not co-occur in the same linguistic context as frequently as bright and red. In the present study, we examined how well MSCs were predicted by an embodied model (switch / no-switch between perceptual modalities) versus a linguistic model (switch / no-switch between linguistic distributional clusters), in behavioural (RT) and continuous event-related EEG potentials (ERP) paradigms. The behavioural data supported the linguistic model in explaining MSCs and found MSCs to be moderated by the target modality, the ERPs showed that linguistic distributional pattern played a crucial role in the neural activations of MSCs. What used to be found as a result of perceptual switching (e.g., "early N400" effect) could be better explained by the linguistic model. The embodied component was activated later than the linguistic component, accounting for activations associated with semantic representation (typically in N400 area). Later during processing, both components were active for decision making (often manifested as LPC).