It is an open question how animals find food in dynamic natural environments where they possess little or no knowledge of where resources are located. Foraging theory predicts that in environments with sparsely distributed target resources, where forager knowledge about resources' locations is incomplete, Lévy flight movements optimize the success of random searches. However, the putative success of Lévy foraging has been demonstrated only in model simulations. Here, we use high-temporal-resolution Global Positioning System (GPS) tracking of wandering (Diomedea exulans) and black-browed albatrosses (Thalassarche melanophrys) with simultaneous recording of prey captures, to show that both species exhibit Lévy and Brownian movement patterns. We find that total prey masses captured by wandering albatrosses during Lévy movements exceed daily energy requirements by nearly fourfold, and approached yields by Brownian movements in other habitats. These results, together with our reanalysis of previously published albatross data, overturn the notion that albatrosses do not exhibit Lévy patterns during foraging, and demonstrate that Lévy flights of predators in dynamic natural environments present a beneficial alternative strategy to simple, spatially intensive behaviors. Our findings add support to the possibility that biological Lévy flight may have naturally evolved as a search strategy in response to sparse resources and scant information.optimal foraging | organism | predator-prey | telemetry | evolution T heoretically, in situations where animals possess limited or no information on the whereabouts of resources, a specialized random walk known as a Lévy flight can yield encounters with sparsely and randomly distributed targets (e.g., prey) more efficiently than random walks such as Brownian motion (1, 2), which are efficient where prey is abundant (3) and probably more predictable (4, 5). Lévy flight, in which movement displacements (steps) are drawn from a probability distribution with a powerlaw tail (a Pareto-Lévy distribution), describes a search pattern composed of many small-step 'walk clusters' interspersed by longer relocations. This pattern is repeated across all scales, such that PðlÞ ∼ l −μ , with 1 < μ ≤ 3 where l is the flight length (move-steplength), and μ the power-law exponent. Simple model simulations of Lévy search generally describe a forager moving along consecutive step lengths drawn from a power law distribution, such that when randomly and sparsely distributed prey is detected within a "sensory" field, the current step length is terminated, the prey is consumed and then a new random direction and step length are selected (3). These Lévy search-model simulations indicate an optimal exponent of μ ≈ 2 for the power-law move-step frequency distribution, leading to searches that increase the probability of a forager encountering new prey patches (1-3). In recent years, Lévy flight or Lévy walk patterns approaching the theoretically optimal value of μ ≈ 2 have been identified in movements of divers...