The Great Recession has challenged the adequacy of existing models to explain key macroeconomic data, and raised the concern that the models might be misspecified. This paper investigates the importance of misspecification in structural models using a novel approach to detect and identify the source of the misspecification, thus guiding researchers in their quest for improving economic models. Our approach formalizes the common practice of adding "shocks" in the model and identifies potential misspecification via forecast error variance decomposition and marginal likelihood analyses. Simulation results based on a small-scale DSGE model demonstrate that the method can correctly identify the source of misspecification. Our empirical results show that state-of-the-art medium-scale New Keynesian DSGE models remain mis-specified, pointing to asset and labor markets as the sources of the misspecification.