A goal of animal movement analysis is to reveal behavioural mechanisms by which organisms utilize complex and variable environments. Statistical analysis of movement data is complicated by the fact that the data are multidimensional, autocorrelated and often marked by error and irregular measurement intervals or gappiness. Furthermore, movement data reflect behaviours that are themselves heterogeneous. Here, we model movement data as a subsampling of a continuous stochastic processes, and introduce the behavioural change point analysis (BCPA), a likelihood-based method that allows for the identification of significant structural changes. The BCPA is robust to gappiness and measurement error, computationally efficient, easy to implement and reveals structure that is otherwise difficult to discern. We apply the analysis to a GPS movement track of a northern fur seal (Callorhinus ursinus), revealing an unexpectedly complex diurnal behavioural profile, and demonstrate its robustness to the greater errors associated with the ARGOS tracking system. By informing empirical interpretation of movement data, we suggest that the BCPA can eventually motivate the development of mechanistic behavioural models.