Integrated Assessment Models (IAMs) are indispensable in the debate over climate change impacts and mitigation policies. Recently these models have incorporated land-based mitigation policies into their analyses. This is important, since land-based emissions account for more than one-quarter of global GHG emissions (Baumert, Herzog, and Pershing 2009), could potentially supply 50% of economically efficient abatement at modest carbon prices, with most of this abatement coming from slowing the rate of agricultural land conversion (Golub et al. 2012). Therefore, projections of agricultural land use are essential inputs to climate change studies. However, the value of such projections hinges on the scientific credibility of the underlying models. And this depends on model validation – an area in which IAMs have been notably lacking to date. Currently, there is great interest in redressing this limitation. However, the challenge is a daunting one, since IAMs seek to integrate not only climate, and the responses of the biophysical system to climate change, but also the economic impacts of such changes. Unlike climate models, economic models must predict human behavior, as well as market interactions between economic agents. In particular, human decision making with respect to land use is context dependent, prone to change over time and poorly understood (Meyfroidt 2012). And even when these relationships are known, there is a lack of global, disaggregated, consistent, time series data for model estimation and evaluation of the full modeling system. In response to this challenge, some modelers have proposed a more targeted approach to validation by focusing on a few key historical developments or ‘stylized facts’ (Schwanitz 2012). This suggests a useful way forward for the IAM community. Without doubt, the most important fact about global land use over the past 50 years has been the tripling of crop production, with only 14% of this total coming at the extensive margin in the form of expansion of total arable lands (Bruinsma 2009). This remarkable accomplishment contributed significantly to moderating land-based emissions (Burney, Davis, and Lobell 2010). Whether or not this historical performance can be replicated in the future is a central question in IAM analysis (Havlik et al. 2012; Wise et al. 2009). Yet, to our knowledge, none of the IAMs currently in use is capable of reproducing this historical experience endogenously. Indeed, it is not uncommon for IAMs to treat crop yields as an exogenous trend (Calvin et al. 2012), thereby pre-determining the answer to this important question. We propose that land-based IAMs be asked to evaluate their models by looking back at this historical experience. In this paper, we illustrate the opportunity and the challenge of undertaking such an historical validation exercise using the SIMPLE model of global crop production. As its name suggests, this framework is designed to be as simple as possible while capturing the major forces at work in determining global crop land use. This makes it a useful test-bed for the design of validation experiments. We test the model’s performance against the historical period: 1961-2006, illustrating what it does well and what it does poorly. Using this 45-year period as our laboratory, and focusing on the dimensions along which the model performs well, we then explore how various model restrictions which appear in the IAM literature alter the model’s historical performance. This serves to highlight which areas of IAM development are likely to be most important from the point of view of global land use change. We conclude with suggestions about how best to advance the state of our knowledge about IAMs by testing these models against history.