We propose and implement a coherent statistical framework for combining theoretical and empirical models of macroeconomic activity. The framework is Bayesian, and enables the formal yet probabilistic incorporation of uncertainty regarding the parameterization of theoretical models. The approach is illustrated using a neoclassical business-cycle model which builds on the Greenwood, Hercowitz, and Huffman (1988) variable-utilization framework to study out-of-sample forecasting of output and investment. The forecasts so produced are comparable with those from a Bayesian vector autoregression.