Nowadays, crosswind stability is a key topic for the homologation of railway vehicles and thus a pivotal boundary condition in their design process. In many countries, including Germany, the safety proof is based on the numerical simulation of the driving behaviour of the vehicle in extreme situations and must necessarily include the aerodynamic and driving performances of the vehicle as well as the wind conditions to be reasonably expected during operation. It follows that the quality of the safety proof depends on the accuracy of the available models. In this respect a deep gap can be observed: on the one hand high accuracy can be reached by multibody simulation in the investigation of the driving dynamics; on the other hand the aerodynamic loads acting on the vehicle, which depend on the vehicle shape and the wind scenario, can be estimated only with poor accuracy. The latter problem is due to the difficulties in the set up of models and the determination of the system parameters because of the complexity of the three dimensional flow around the vehicle and the implicit stochastic nature of atmospheric wind. In the present norms for crosswind stability such modelling uncertainties are usually not considered or are very empirically taken into account by safety factors. In this work some improvement in the modelling of the aerodynamic phenomena and the driving dynamics are firstly introduced. It could be observed that with regard to aerodynamics more complex models than usual are necessary, e.g. to cover unsteady phenomena; on the contrary, simpler models than usual can be sufficient for the analysis of the driving dynamics, allowing, for example, the use of linear system theory. Then, in order to include parametric uncertainty in the safety proof, the conventional risk analysis for crosswind stability has been coupled with methods from reliability analysis. Such methods, which are quite common in structural mechanics and are available in different formulations, lead to the efficient assessment of the risk and thus to a reduction of the safety factors, provided that a statistical description of the uncertainties is available. Even though the provision of such a description is often a challenge, good results can be also achieved on the basis of little available information. Furthermore, sensitivity analysis and optimisation can be reformulated on the basis of the proposed approach, so that reliability analysis can be integrated not only in the safety proof but also in the design process. The discussed methods have been tested on the real case of a German high speed train.