Studies of nuclear proliferation share five serious problems. First, nuclear programs' initiation and completion dates are ambiguous and difficult to code, but findings are rarely subjected to sufficient robustness tests using alternative codings. Second, independent variables overlook important factors such as prestige and bureaucratic power and often use poor proxies for concepts such as the nonproliferation regime. Third, methodologies and data sets should be tightly coupled to empirical questions but are instead often chosen for convenience. Fourth, some findings provide insights already known or believed to be true. Fifth, findings can ignore or gloss over data crucial for policy making and wider debates. This article reviews new quantitative research on nuclear proliferation, noting improved analysis and lingering problems. It highlights the 1999 Kargil war to explore dangers of relying on stock data sets and the need for research on statistical outliers. It concludes with a future research agenda aimed at correcting problems and a cautionary note regarding hasty application of quantitative results to policy making.