||In the context of the expansion of animal tracking and bio-logging, state-space models have been developed with the objective to characterise animals' trajectories and to understand the factors controlling their behaviour. In the fisheries community, the electronic tagging of vessels commonly designated by Vessel Monitoring Systems (VMS) is developing and provides a new insight for the understanding, the analysis and the modelling of the trajectories of vessels and their prospecting behaviour. VMS data are thus a clue for the proper definition of fishing effort which remains a fundamental parameter of tuna stock assessments. In this context, we used the VMS (recording of hourly positions) of the French tropical tuna purse-seiners operating in the Indian Ocean to characterise three types of movement (states) on the VMS trajectories (stillness, tracking, and cruising). Based on empirical evidences, and on the regular frequency of VMS acquisition, this was achieved by the development of a Bayesian Hidden Markov model for the speeds and turning angles derived from the hourly steps of the trajectories. In a second phase, states were related to activities disentangling stillness into fishing or stop at sea. Finally the quality of the model performances was rigorously quantified thanks to observers' data. Confronting model prediction and true activities allowed estimating that 10% of the hourly steps were misclassified. The assumptions and model' choices are discussed, highlighting the fact that VMS data and observers' data having different time resolutions, the effective use of validating data was troublesome. However, without validation, these analyses remain speculative. The validation part of this work represents an important step for the operational use of state-space models in ecology in the broad sense (predators' tracking data, e.g. birds or mammals trajectories).