Shaking tables provide one means of obtaining data from physical models of geotechnical systems under simulated earthquake loading. While there may be challenges in ensuring exact scaling of all aspects of a geotechnical system between prototype and model, the large size of the shaking table models does give a good basis for validation of numerical modelling strategies. A shear stack provides a means of creating a mass of soil which can approximate the free field of the real world. Observations of the performance of the Bristol shear stack are presented and compared qualitatively with patterns of behaviour observed in laboratory simple shear tests. The results of some tests on model gravity retaining walls performed using the shear stack are used to show the relevance of simple models of geotechnical system performance and to illustrate the difficulty of correlating system performance with individual parameters of input time histories.