With rising sea levels at the end of the Pleistocene, land-bridge or continental islands were formed around the world. Many of these islands have been extensively studied from a biogeographical perspective, particularly in terms of impacts of island creation on terrestrial vertebrates. However, a majority of studies rely on contemporary faunal distributions rather than fossil data. Here, we present archaeological findings from the island of Zanzibar (also known as Unguja) off the eastern African coast, to provide a temporal perspective on island biogeography. The site of Kuumbi Cave, excavated by multiple teams since 2005, has revealed the longest cultural and faunal record for any eastern African island. This record extends to the Late Pleistocene, when Zanzibar was part of the mainland, and attests to the extirpation of large mainland mammals in the millennia after the island became separated. We draw on modeling and sedimentary data to examine the process by which Zanzibar was most recently separated from the mainland, providing the first systematic insights into the nature and chronology of this process. We subsequently investigate the cultural and faunal record from Kuumbi Cave, which provides at least five key temporal windows into human activities and faunal presence: two at the end of the Last Glacial Maximum (LGM), one during the period of post-LGM rapid sea level rise and island formation, and two in the late Holocene (Middle Iron Age and Late Iron Age). This record demonstrates the presence of large mammals during the period of island formation, and their severe reduction or disappearance in the Kuumbi Cave sequence by the late Holocene. While various limitations, including discontinuity in the sequence, problematize attempts to clearly attribute defaunation to anthropogenic or island biogeographic processes, Kuumbi Cave offers an unprecedented opportunity to examine post-Pleistocene island formation and its long-term consequences for human and animal communities.