A rich literature has identified a number of important drivers of nuclear proliferation. Most of this work, however, treats the determinants of proliferation as constant over the entire nuclear age—the factors leading to proliferation are assumed to be the same in 2010 as they were in 1945. But there are reasons to suspect that the drivers of proliferation have changed over this time: nuclear technology is easier to come by, the global strategic environment has shifted, and the nuclear nonproliferation regime has come into being. To examine how the drivers of nuclear proliferation have changed over time, I adapt a cross-validation technique frequently used in the machine learning literature. I create a rolling window of training data with which statistical models of proliferation are built, and I then test the predictive power of these models against data from other time periods. The result of this analysis is a temporal map of how the determinants of proliferation have changed over time. My findings suggest that the underlying dynamics of nuclear proliferation have indeed changed over time, with important implications both for the literature on nuclear proliferation and for policymakers interested in limiting the future spread of nuclear weapons.