Path integration (dead reckoning) is a computational process by which an animal can keep track of its position relative to some starting point by relying on self-generated information collected en route. The purpose of this paper is two-fold: first, to review four important models of path integration, analysing how the different models have evolved as a result of the authors' particular focus; second, to assess what they have brought to the understanding of path integration. The more specific the model, the less it will be possible to apply it to different situations; the more general its nature, the less adequately it will represent the details of the behaviour to be modelled. In this context, the paper discusses how the authors envisioned the role of homing errors. The usefulness and limits of the different perspectives is discussed, in particular how well the models fit with observations. Finally, ways of improving and developing the modelling process are proposed.