Previous neuroimaging studies have reported that learning multi-sensory associations involves the superior temporal regions (Tanabe et al., 2005). However, the neural mechanisms underlying the retrieval of multi-sensory associations were unclear. This functional MRI (fMRI) study investigated brain activations during the retrieval of multi-sensory associations. Eighteen right-handed college-aged Japanese participants learned associations between meaningless pictures and words (Vw), meaningless sounds and words (Aw), and meaningless sounds and visual words (W). During fMRI scanning, participants were presented with old and new words and were required to judge whether the words were included in the conditions of Vw, Aw, W or New. We found that the left superior temporal region showed greater activity during the retrieval of words learned in Aw than in Vw, whereas no region showed greater activity for the Vw condition versus the Aw condition (k > 10, p < .001, uncorrected). Taken together, the left superior temporal region could play an essential role in the retrieval process of auditory-word associations.