This paper describes the development of a program of project-based learning in Calculus courses at a large urban research university. In this program, students developed research projects in consultation with a faculty advisor in their major, and supervised by their calculus instructors. Students wrote up their projects in a prescribed format based on scientific publishing norms. These write-ups were used as part of the basis for their course grades, typically as replacements for their final exams. Because this program involved a large number of students over several years, we were able to evaluate whether there were differences in how well students performed in subsequent courses (both mathematics and non-math courses required for engineering majors) (a) according to whether students were in sections offering projects, in concurrent non-project sections, or historical controls (prior to the project), and (b) among those in sections offering projects, according to whether they completed a project. We used as endpoints the pass and withdrawal rates in both the calculus courses themselves and in the follow-up courses. In Calculus II, students in project-option sections performed better than those in non-project-option sections, and this was true for every social subdivision examined (sex and race/ethnicity). There was a similar but smaller effect for Calculus III, which may thus be due to random fluctuations. Students submitting projects were more likely to go on to follow-up courses. Given enrollment in follow-up courses, there was no detectable performance difference between those who had and those who had not submitted projects; greater throughput does occur among those submitting projects, but this is associated with their greater enrollment in follow-up courses. In implementing this program, we encountered challenges in dealing with students, with the institution, and with individual faculty; we discuss the key challenges here.