For railway positioning solutions based on GNSS (Global Navigation Satellite Systems) like the GPS (Global Positioning System) or the future Galileo, a generic model is impossible to create in regards to the signal degradations in the atmosphere, the multipath effects caused by receiver near-environment, the multitude of environment configurations crossed by the train and the weak feedback of these technologies for estimated failure rates. To compensate the weakness of GNSS, it must be hybridised with other sensors to determine a position sufficiently accurate for an use in safety applications. A multitude of information sources is available about the train position. Only one position is possible. In consequence, a fusion step is necessary to combine all these sources of position. This raises some questions: Why the technology hybridisation is interesting to provide a accurate position? How the influences of sensor errors can affect the system output? Which sensors combination is the most efficient in regards to RAMS (Reliability, Availability, Maintainability and Safety) analyses required in railway safety standards? This paper proposes to focus on this last question with an analysis of different sensor architectures in order to understand how errors (propagation of failures) of one or several sensors can affect the entire positioning system. To answer to this question, a causal analysis is led based on the sensor behaviours.