We compare travel distance estimation from path integration during walking with path integration from visual flow. For visually simulated self-movement humans typically underestimate travel distance, which can be explained by leaky path integration. The amount of leak, i.e. the underestimation, is determined by the length of the path. For visually simulated movements along curved paths that veer left and right around a central forward direction estimates of the start-to-end distance decrease as the veering, i.e. the path length increases. Leaky path integration for visual travel distance estimation thus takes place along the actually traversed path even when a straight beeline distance is calculated. We studied whether the same leaky path integration occurs during real self-motion when vestibular and proprioceptive cues are available instead of vision. Sixteen subjects walked blindfolded from a starting point to targets 20, 30 or 40m away, guided by an experimenter. They walked either along a straight line, or along paths that deviated first to the right and then to the left (or vice versa) before they reached the end point. This increased the path length by 5%, 10%, 20% or 30%. Subjects then gave a verbal estimate of their beeline distance from the starting point. Like in the visually simulated case, distance estimates for the same start-to-end distance of 40m dropped as the path length increased, consistent with the prediction of the leaky integration model. We conclude that travel distance estimation is similar for visual and for vestibular/proprioceptive cues.