Abstract Probabilistic approaches encourage a healty scepticism towards assumptions in modelling and design as well as towards the test data. This paper reflects on the various probabilistic approaches and recommends Monte Carlo simulation as the one with the most promise of success in system reliability studies. This technique implemented in the form of interactive and tolerant educational and applications programs overcomes effectively and efficiently the limitations of the analytical approaches. An example of a retaining wall is used to illustrate the relative merits of the methods and to validate the conclusions drawn. It is argued that even an approximate answer to a right question is far more reliable than a notionally ls/exact’ one to a wrong question; the latter being the result of gross simplifications made to make the problem analytically tractable.