The risk of predation is a strong force shaping many aspects of animal behaviour. Early detection and efficient avoidance strategies not only help prey to survive, but also limit the negative impact predation has on other aspects of their lives. The type of antipredator strategy an animal uses must depend on its ability to collect accurate information on the risks to which it is exposed. However, models attempting to predict when a prey animal should escape from an approaching predator often assume that the prey has accurate information on a predator's distance and direction of approach. To test whether such models could be applied to a prey animal with restricted sensory capabilities, I explored the predator avoidance behaviour of the fiddler crab Uca vomeris by approaching crabs with small dummies intended to mimic a hunting tern. The crabs responded strongly and reliably to the simple dummies. They were clearly sensitive to risk and responded more often and earlier, the further away they were from their refuge. The probability of response was most strongly influenced by how directly and therefore how closely the dummies approached the crabs and reached 100% for very direct approaches. Surprisingly, however, the crabs responded later when the dummies approached them more directly. I argue that this counterintuitive result reflects a lack of reliable information on the predator's distance and movement relative to the crab and its refuge, because like many small animals, crabs cannot measure the distance to a predator. I conclude that general models attempting to predict when animals should respond to an approaching predator need to incorporate the information that prey animals have available at the time of response.