This quantitative study was designed to investigate teacher preparation as an interactive system and examine whether individual approaches to preparation are associated with differential retention benefits across different teacher subgroups. Drawing on longitudinal data on mathematics teachers who entered teaching through the New York City Teaching Fellows (NYCTF) program, the study focused specifically on the first-school retention of two policy-relevant teacher subgroups, namely, (1) the graduates of very selective colleges and (2) Black and Latinx community-insiders. Whereas the latter all attended city high schools, the former generally did not. In part because the field lacks a well-substantiated theory of the relationship between teacher initial preparation and their retention, we drew on intersectionality theory and incorporated two-way interaction effects in quantitative models of the teachers’ first-school retention. We found significant relationships between NYCTF’s initial training and the teachers’ retention and, further, that single approaches to initial training appeared to bestow different retention benefits to different teacher subgroups. We also found that the Black and Latinx community-insiders exhibited significantly higher rates of first-school retention than the elite college graduates and, for that matter, other NYCTF mathematics teachers. This result has clear implications for teacher recruitment and training and, in particular, the promise of developing community-based mathematics teachers who resemble the students they teach.