Previous studies on visual working memory (VWM) have primarily investigated memory for an array presented for a single moment. Here, we examined VWM for two arrays separated by a 1,100-msec interval. We focused on the allocation of VWM capacity to the two arrays as a function of dynamic events inserted between them. During the interval, irrelevant dots moved to form three types of motion: (1) coherent (apparent) motion that connected Arrays 1 and 2, (2) jumpy motion, or (3) coherent motion with two disconnected segments. Results showed that VWM for Array 2's locations was better than for Array 1's, especially when the arrays were connected by coherent motion. We suggest that coherent motion between two temporally disparate arrays connects the arrays into a single visual event. VWM is then biased toward remembering the more recent state of the event.