It is often assumed that navigation implies the use, by animals, of landmarks indicating the location of the goal. However, many animals (including humans) are able to return to the starting point of a journey, or to other goal sites, by relying on self-motion cues only. This process is known as path integration, and it allows an agent to calculate a route without making use of landmarks. We review the current literature on path integration and its interaction with external, location-based cues. Special importance is given to the correlation between observable behavior and the activity pattern of particular neural cell populations that implement the internal representation of space. In mammals, the latter may well be the first high-level cognitive representation to be understood at the neural level.