Combat and operational stress control (COSC) surveys guide allocation of high-demand, low-quantity mental health assets to support combat-deployed U.S. forces. The current article describes an innovative application of machine learning, decision tree analysis, to predict unit-level risk for combat mental health outcomes like posttraumatic stress disorder (PTSD). The initial algorithm was developed from large population-based COSC surveys conducted in 2007/2008 in Iraq and Afghanistan. The algorithm was validated in a separate sample of COSC surveys collected in Afghanistan in 2010. Using the applied field standard for high-risk units (i.e., 10% or more of the unit screening at risk for PTSD), the decision tree algorithm correctly identified 100% of units considered high risk for PTSD in the validation sample, while only misclassifying 10% (3 of 31 units) in the independent 2010 sample. This article provides a template by which future efforts to enhance COSC can be aided by iterative approaches to analyzing “big” behavioral health data sets.