Prioritizing efforts for conserving rare and threatened species with limited past data and lacking population estimates is predicated on robust assessments of their occupancy rates. This is particularly challenging for elusive, long-lived and wide- ranging marine mammals. In this paper we estimate trends in long-term (over 50 years) occupancy, persistence and extinction of a vulnerable and data-poor dugong (Dugong dugon) population across multiple seagrass meadows in the Andaman and Nicobar archipelago (India). For this we use hierarchical Bayesian dynamic occupancy models accounting for false negatives (detection probability,1), persistence and extinction, to two datasets: a) fragmentary long-term occurrence records from multiple sources (1959–2004, n = 40 locations), and b) systematic detection/non-detection data from current surveys (2010–2012, n = 57). Dugong occupancy across the archipelago declined by 60% (from 0.45 to 0.18) over the last 20 years and present distribution was largely restricted to sheltered bays and channels with seagrass meadows dominated by Halophila and Halodule sp. Dugongs were not found in patchy meadows with low seagrass cover. In general, seagrass habitat availability was not limiting for dugong occupancy, suggesting that anthropogenic factors such as entanglement in gillnets and direct hunting may have led to local extinction of dugongs from locations where extensive seagrass meadows still thrive. Effective management of these remnant dugong populations will require a multi-pronged approach, involving 1) protection of areas where dugongs still persist, 2) monitoring of seagrass habitats that dugongs could recolonize, 3) reducing gillnet use in areas used by dugongs, and 4) engaging with indigenous/settler communities to reduce impacts of hunting.