Desert ants (Cataglyphis sp.) monitor their position relative to the nest using a form of dead reckoning    known as path integration (PI) . They do this with a sun compass and an odometer to update an accumulator that records their current position . Ants can use PI to return to the nest  . Here, we report that desert ants, like honeybees  and hamsters , can also use PI to approach a previously visited food source. To navigate to a goal using only PI information, a forager must recall a previous state of the accumulator specifying the goal, and compare it with the accumulator's current state . The comparison - essentially vector subtraction - gives the direction to the goal. This whole process, which we call vector navigation, was found to be calibrated at recognised sites, such as the nest and a familiar feeder, throughout the life of a forager. If a forager was trained around a one-way circuit in which the result of PI on the return route did not match the result on the outward route, calibration caused the ant's trajectories to be misdirected. We propose a model of vector navigation to suggest how calibration could produce such trajectories.