The US military attempts to recruit personnel from all demographic, economic, and cultural segments of the nations youth population. However, since the beginning of the all-volunteer military over 40 years ago, certain regions of the nationespecially the Southhave supplied disproportionate numbers of recruits relative to the size of their youth population. This paper probes more deeply than just regional identification and summary economic indicators such as a states youth unemployment rate. The supporting research uses machine learning methods (in particular, regression trees) to identify the demographic, economic, and cultural features of individual counties (or small collections of contiguous counties) that are most influential in predicting non-prior-service enlisted accessions. Among the most influential predictors are various measures of the proportion of veterans in the countys population. Other important predictors are the percentage of people attending college, the percentage who have completed some college (but not a bachelors degree), the fraction of high school students available to take the Armed Services Vocational Aptitude Battery, and the prevalence of Junior Reserve Officers Training Corps programs.