There is widespread concern, both in the private and public sectors, about perceived declines in U.S. college graduates in STEM fields. In our sample, the proportion of science majors has remained steady over the sample period; however, the number entering our college intending to major in STEM fields has fallen. In this paper we use administrative data from the graduating classes of 2001-2009, roughly 5000 graduates, from a northeastern liberal arts college to model the progression of students through STEM majors. A series of selection models predicts the choice of whether to take a second course in the department, conditional upon having taken a first course. This choice is modeled as a function of pre-college characteristics and preferences, characteristics of the student, the course, the professor, the peers in the course, and the grade received in the course. Using the selected sample that progresses to a second course, the choice to progress to a third is modeled conditional on having taken the second. The covariates in these models are similar to those in the first stage. Models are estimated for the Biology, Chemistry, Computer Science, Geology, Mathematics, Physics, and Psychology majors. Results suggest that gender effects are important, both in terms of the influence of the absolute and relative grades received, and in some cases in terms of the peers in the course and the gender of the instructor. The intended major (as reported on the admissions application) is a strong indicator of the likelihood of taking initial courses in a discipline and progression to a second course. AP credits are also strongly correlated to taking a first course, but diminish in the more selected samples. Grades and pre-collegiate intended major, have the most consistent and important influence on the decision to progress in a STEM major. When comparing across men and women, grades play a more important role in men’s decision-making while preferences play a bigger role in women’s choices.