Economists and psychologists have devised numerous instruments to measure time preferences and have generated a rich literature examining the extent to which time preferences predict important outcomes; however, we still do not know which measures work best. With the help of a large sample of non-student participants and administrative data on outcomes, we gather four different time preference measures and test the extent to which they predict both on their own and when they are all forced to compete head-to-head. Our results suggest that the now familiar (β, δ) formulation of present bias and exponential discounting predicts best, especially when both parameters are used.