The development of research skills and scientific reasoning underpins the mission of graduate education in science, technology, engineering and mathematics (STEM) fields, yet our understanding of this process is mainly drawn from self-report and faculty survey data. In this study, we empirically investigate the pattern of research skill development using STEM graduate students' written research proposals. Analyses of proposal performance data suggest a potential developmental trajectory of research skills, in which the ability to effectively situate work in context using primary literature, and to generate testable hypotheses, emerge early in students' careers, while other skills, such as data analysis and forming conclusions from data, appear to develop later. We discuss these findings in relation to threshold concepts theory, a framework which posits that intellectual growth occurs in transformative leaps rather than a gradual progression, especially as it applies to graduate student research skill development. (Contains 5 tables and 4 figures.)