Soldiers who exit the Army before the end of their first term of service represent lost investments in terms of both direct costs (e.g., training costs) and indirect costs (e.g., force instability). The Army would benefit from pre-enlistment information that could identify those individuals most likely to complete their first term. This paper examines the relationship between first-term military attrition, information currently available to the Army (e.g., high school diploma graduate status), and three non-cognitive measures developed and administered during the Army's Project A/Building the Career Force research program (Campbell and Zook, 1990a; Campbell and Zook, 1992). Proportional hazards regression models were estimated for 4 job groups (roughly 49,000 first-term soldiers) using hierarchical and empirical "best-model" approaches. A battery of predictors is proposed for predicting attrition in all four job groups. The Army can improve prediction of attrition by gathering biodata and temperament information prior to enlistment and by better using information they currently possess. Nine figures and 16 tables present study information. (Contains 26 references.) (Author/SLD)