Serial dependence is a phenomenon in which current perception is attracted to the immediately preceding perception and is thought to reflect a mechanism for stabilizing perception. It has been demonstrated across a variety of stimuli and has also been observed in the numerosity perception. A previous study suggested that cross-modal serial dependence in numerosity perception from audition to vision did not occur; however, differences in the stimulus presentation format might have prevented serial dependence from emerging. Therefore, we used a standardized temporal presentation format. As a result, we observed bidirectional cross-modal serial dependence between audition and vision. However, the effects in each direction were not consistent within individuals. These findings provide important insights into the mechanisms underlying serial dependence, indicating that both higher-level processing, such as abstract numerical representation, and lower-level processing, such as auditory and visual cortex involvement, are engaged in cross-modal serial dependence in numerosity perception.