Recent evidence suggests that the default options implicit in economic choices (e.g., 401(k) savingsby white-collar workers) have extraordinarily large effects on decision-making. This study presentsa field experiment that evaluates the effect of defaults on savings among a highly policy-relevant population:low-income tax filers. In the control condition, tax filers could choose (i.e., opt in) to receive someof their federal tax refund in the form of U.S. Savings Bonds. In the treatment condition, a fractionof the tax refund was automatically directed to U.S. Savings Bonds unless tax filers actively choseanother allocation. We find that the opt-out default had no impact on savings behavior. Furthermore,our treatment estimate is sufficiently precise to reject effects as small as one-fifth of the participationeffects found in the 401(k) literature. Ancillary evidence suggests that this "nudge" was ineffectivein part because the low-income tax filers in our study had targeted plans to spend their refunds. Theseresults suggest that choice architecture based on defaults may be less effective in certain policy-relevantsettings, particularly where intentions are strong.