Computable General Equilibrium (CGE) models are commonly used for global agricultural market analysis. However, concerns are sometimes raised about the quality of their output since key parameters may not be econometrically estimated and little emphasis is generally given to model assessment. This article addresses the latter issue by developing an approach to validating CGE models based on the ability to reproduce observed price volatility in agricultural markets. We show how patterns in the deviations between model predictions and validation criteria can be used to identify the weak points of a model and guide development of improved specifications with firmer empirical foundations.