An entropy-based approach for flow loss characterization with computational fluid dynamics (CFD) is presented. Unlike past methods of global loss characterization, this article outlines a new approach for predicting local losses of available energy. The local entropy generation provides information regarding the spatial distribution of mechanical energy loss, which can be used to systematically optimize thermofluid systems. An application representing subsonic flow through a diffuser is investigated. The main parameter under consideration is the desired inlet expansion angle, which yields the minimum entropy generation at a specified Reynolds number and inlet flow condition. The numerical results indicate that the entropy-based approach offers a new way of establishing the optimal diffuser configuration exhibiting minimal flow losses. By successfully predicting the local flow irreversibilities, re-design efforts can be more carefully focused on specific regions of highest entropy production. Copyright (c) 2005 John Wiley & Sons, Ltd.