Effective management of wild populations of animals relies on having accurate methods for estimating their densities from field surveys. Surveying sharks and other highly mobile marine predators is particularly difficult, requiring specialised methods that can entail a range of potential biases. Recent simulation studies have suggested that non-instantaneous visual and video survey methods may overestimate densities of highly mobile fishes. A non-instantaneous method is one where each point in space is surveyed for a non-negligible period of time, rather than in an instantaneous ‘snapshot’, allowing animals to move into the survey space during the survey period. We conducted instantaneous and non-instantaneous surveys of sharks (Sphyrna lewini and Carcharhinus galapagensis) around Darwin and Wolf Islands (Galápagos Islands, Ecuador) using diver-operated stereo-video (stereo-DOV) systems with towed GPS. We provide the first empirical evidence of non-instantaneous bias and show that the degree of bias is negatively related to diver swimming speed. At a speed of 40 m min-1, the non-instantaneous method recorded twice as many sharks as the instantaneous method. The bias may be as high as 10 fold at slow speeds (23 m min-1) and negligible at very fast speeds (65 m min-1). Moreover, we used GPS data to produce fine-scale distribution maps, revealing high densities in areas of strong currents along the south-eastern sides of both islands. We recommend the use of stereo-DOV systems with GPS and fast swimming speeds for surveying highly mobile sharks, to produce more realistic estimates of densities on which to base management of these species.