Established analyses of labour market commuting are based on random choice models and gravity type models. In these models generalised transport costs are formulated as exponential or log-linear distance dependent functions. This paper presents empirical observations, which imply that time distances influence the commuting behaviour in a non-linear way, such that the time sensitivity is much lower for very short and long distances, whereas intermediate distances display a high time sensitivity. This is explained in a model that is parameterised and estimated. The results are important for understanding and predicting commuter behaviour. It also helps to delineate space, as in classical traditions, into local, intra-regional and extra-regional space.