In earthquake engineering, structural models are validated by performing a time history analysis and comparing its maximum to the maximum response obtained by a shake table test. It has been shown that this is a sufficient (but not a necessary) precondition to accept a numerical model. Numerical models can fail to predict the planar rocking response of a rigid block, but may succeed in predicting the statistics of the response to an ensemble of ground motions. As seismic response is inherently stochastic, comparison of the statistics of the numerically simulated response to the statistics of the experimentally obtained benchmark response for the same ensemble of earthquake excitation is a sufficient (and easier to pass) model validation test. This article describes the publicly available data of a set of 12 free rocking vibration and 115 shake table tests of six three-dimensional rocking and sliding columns, designed at ETH Zurich and performed at EQUALS Laboratory, University of Bristol. The data can be used to statistically validate different approaches that aim to model three-dimensional rocking structures.