This paper analyses and compares two probabilistic models described in the literature concerning pipe breakage in a water distribution system. The first model was proposed by Mailhot et al. (Mailhot, A., et al., 2000. Modeling the evolution of the structural state of water pipe networks with brief recorded pipe break histories: methodology and application. Water Resources Research, 36 (10), 3053–3062), whereas the second was proposed by Le Gat and Eisenbeis (Le Gat, Y. and Eisenbeis, P., 2000. Using maintenance record to forecast future failures in water networks. Urban Water, 2, 173–181). Both models are presented by the respective authors as being parameterisable even using data relating to a time window that is not very broad; they were thus applied to a case study on the water distribution system serving the western part of the province of Ferrara, for which the available breakage data cover a time span of nearly 7 years and comprise a total of around 3500 breakages. This data sample is one of the richest and most complete in Italy. A preliminary statistical analysis of these data and of the main characteristics of the transmission and distribution system resulted in the selection of two different parameterisation approaches for the two models, in which the data are stratified, respectively, according to materials and the periods in which the pipes were laid. With each of the two models, tests were then carried out to assess (a) their ability to reproduce, in a probabilistic sense, the number of breakages and the number of broken pipes recorded over the observation time window; (b) how they behave in forecasting the number of breakages in the years following the observation period; and (c) how they behave in quantifying the benefits to be derived, in terms of reducing breakages, from the planning of pipe rehabilitation initiatives. An analysis of the results of these tests showed that the two models perform similarly in modelling the number of breakages and the number of broken pipes recorded during the observation period. However, as regards their behaviour in forecasting the number of future breakages and in quantifying the benefits to be derived from rehabilitation, the model proposed by Le Gat and Eisenbeis (2000) appears to be more reliable and less sensitive to the type of data stratification than the model proposed by Mailhot et al. (2000). This is due to the structure of the model of Le Gat and Eisenbeis (2000), which enables optimal exploitation of the abundant information available for a large-sized water distribution system, including diameters, lengths, etc., whereas the model proposed by Mailhot et al. (2000) does not use this type of information and takes into account solely the pipe installation period.