Résumé : Proton therapy uses proton beams with energies of 70 – 230 MeV to treat cancerous tumours very effectively, while preserving surrounding healthy tissues as much as possible. During nuclear interactions of these protons with matter, secondary neutrons can be produced. These neutrons can have energies ranging up to the maximum energy of the protons and can thus be particularly difficult to attenuate. In fact, the rooms of a proton therapy facility are generally surrounded by concrete walls of at least ~2 m in thickness, in order to protect the members of the staff and the public from the stray radiation. Today, the design of the shielding walls is generally based on Monte Carlo simulations. Amongst the numerous parameters on which these simulations depend, some are difficult to control and are therefore selected in a conservative manner. Despite these conservative choices, it remains important to carry out accurate neutron dose measurements inside proton therapy facilities, in order to assess the effectiveness of the shielding and the conservativeness of the simulations. There are, however, very few studies in literature which focus on the comparison of such simulations with neutron measurements performed outside the shielding in proton therapy facilities. Moreover, the published measurements were not necessarily acquired with detectors that possess a good sensitivity to neutrons with energies above 20 MeV, while these neutrons actually give an important contribution to the total dose outside the shielding. A first part of this work was dedicated to the study of the energy response function of the WENDI-2, a rem meter that possesses a good sensitivity to neutrons of more than 20 MeV. The WENDI-2 response function was simulated using the Monte Carlo code MCNPX and validation measurements were carried out with 252Cf and AmBe sources as well as high-energy quasi-monoenergetic neutron beams. Then, WENDI-2 measurements were acquired inside and outside four rooms of the proton therapy facility of Essen (Germany). MCNPX simulations, based on the same conservative choices as the original shielding design simulations, were carried out to calculate the neutron spectra and WENDI-2 responses in the measurement positions. A relatively good agreement between the simulations and the measurements was obtained in front of the shielding, whereas overestimates by at least a factor of 2 were obtained for the simulated responses outside the shielding. This confirmed the conservativeness of the simulations with respect to the neutron fluxes transmitted through the walls. Two studies were then carried out to assess the sensitivity of the MCNPX simulations to the defined concrete composition and the selected physics models for proton and neutron interactions above 150 MeV. Both aspects were found to have a significant impact on the simulated neutron doses outside the shielding. Finally, the WENDI-2 responses measured outside the fixed-beam treatment room were also compared to measurements acquired with an extended-range Bonner Sphere Spectrometer and a tissue-equivalent proportional counter. A satisfactory agreement was obtained between the results of the three measurement techniques.