In a recent article in the Journal of Politics, Golder, Golder, and Siegel (2012) argue that models of government formation should be rebuilt “from the ground up.” They propose to do so with a “zero-intelligence” model of government formation. They claim that this model makes no theoretical assumptions beyond the requirement that a potential government, to be chosen, must be preferred by all its members and a legislative majority to the incumbent administration. They also claim that, empirically, their model does significantly better than existing models in predicting formation outcomes. We disagree with both claims. Theoretically, their model is unrestrictive in terms of its institutional assumptions, but it imposes a highly implausible behavioral assumption that drives the key results. Empirically, their assessment of the performance of the zero-intelligence model turns on data that are of limited relevance in testing coalition theories. We demonstrate that the predictions of the zero-intelligence model are no more accurate than random guesses, in stark contrast to the predictions of well-established approaches in traditional coalition research. We conclude that scholars would be ill-advised to dismiss traditional approaches in favor of the approach advanced by Golder, Golder, and Siegel.