Predictive modelling in the field of dependability, such as repair cost modelling, is usually based on the failure intensity function of repairable systems. The experience feedback data needed to estimate the failure intensity function is, however, seldom available due to the unique character of the systems. Thus approaches that can capture the tacit knowledge of designers, operators and maintenance personnel and transform this to the mathematical format required in the predictive models, are needed. A Bayesian statistical approach is presented, which yields posterior distributions of the parameters of the Power Law and the Log-Linear intensity functions, which are used to model the trend in the data observed. Model-checking criteria are presented according to which the relative performance of the selected functions can be assessed. Numerical examples are also given.