This paper describes the techniques employed in a probabilistic assessment of a riveted truss railway bridge. The probability-based classification of the structure serves as an example of how probability-based assessment of railway bridges can be applied to reduce maintenance costs through avoidance of unnecessary repair/rehabilitation and/or to optimise those repairs which are shown to be necessary. Probabilistic modelling of the critical limit states is presented for both the elements and the riveted joints of the structure. The statistical techniques employed in modeling the train loads are presented, with modelling of the train load extreme value distributions based upon the number of wagons loading the critical length of an influence line for the considered element/joint. Modelling of dynamic amplification of specific static sectional forces is performed as a function of the local or global influence length for the element or joint. Thereby the significant conservatism found to be inherent in the deterministic assessment was avoided. The overall aim of the analysis was to achieve a higher load rating for element/joints of the structure than those resulting from the deterministic assessment. The sensitivity of the reliability index to the modelled stochastic variables is presented. Ultimately, the economic benefits to bridge owners/managers of performing a probabilistic assessment are apparent from the results, which provided a higher load rating for the critical elements/joints of the steel arch bridge than those achieved through deterministic assessment. The results of the probabilistic assessment were not able to demonstrate sufficient capacity in all cases, but in those cases, as demonstrated in the paper, probabilistic modelling was used to plan the optimal repair strategy.