This paper addresses the problem of finding the appropriate method for conducting empirical validation in agent-based (AB) models, which is often regarded as the Achilles' heel of the AB approach to economic modelling. The paper has two objectives. First, to identify key issues facing AB economists engaged in empirical validation. Second, to critically appraise the extent to which alternative approaches deal with these issues. We identify a first set of issues that are common to both AB and neoclassical modellers and a second set of issues which are specific to AB modellers. This second set of issues is captured in a novel taxonomy, which takes into consideration the nature of the object under study, the goal of the analysis, the nature of the modelling assumptions, and the methodology of the analysis. Having identified the nature and causes of heterogeneity in empirical validation, we examine three important approaches to validation that have been developed in AB economics: indirect calibration, the Werker-Brenner approach, and the history-friendly approach. We also discuss a set of open questions within empirical validation. These include the trade-off between empirical support and tractability of findings, the issue of over-parameterisation, unconditional objects, counterfactuals, and the non-neutrality of data.