Both physical and numerical modelling have their advantages and disadvantages in predicting the behaviour of structures on liquefiable soils under earthquake loading. Whilst it can be seen that as a designer, numerical techniques offer the quickest and cheapest route to a safe design, in order to have confidence in the results gained from any numerical code, its results must be checked against those from a physical prototype, whether that is a centrifuge test or a full-scale experiment. In order for this test to be unbiased, numerical analyses should be carried out before the tests in a so-called Class A or ‘blind’ prediction of the behaviour.

This paper will discuss the results of Class A numerical predictions of the behaviour of a shallow bridge foundation on liquefiable soil together with the results later obtained from a dynamic centrifuge model with the same geometry. It will be shown that when a properly calibrated numerical model is used, a very good prediction of the physical model behaviour can be achieved with currently available dynamic FE codes.