Animal motion is a fairly complex phenomenon as it is driven by objectives that are both local (e.g. the wish to travel under a deciduous land-cover) and long term (e.g. going towards a location rich in nutriments). It is also a multi-state process: a feeding animal and an animal traveling between feeding sites move in different ways. For large mammals, data on animal movement is collected with GPS markers given the animal positions at fixed time intervals. This is combined to environment data providing information such as land cover and forest road localisations through a GIS. This talk presents statistical models relating the 2D vector giving the motion between time t-1 and time t to environmental characteristics. A large discrete choice model is proposed to account simultaneously for the animal local and long term objectives and for the multi state nature of the movement. The model is fitted using the EM algorithm where the underlying state is estimated using a filtering smoothing technique. An example dealing with the motion of a bison over a 7 month period is used to illustrate the methodology. This is joint work with A. Nicosia, T. Duchesne and D. Fortin.