We use the random assignment of a private Wisconsin need-based grant to estimate the impacts of financial aid on college persistence among Pell Grant recipients at 13 public universities over three years. For equity and efficiency reasons, governments use conditional cash transfers to reduce the relationship between family income and college attainment, but prior research suggests that financial aid generates only modest positive effects. This is the first experimental study of a program resembling the longstanding federal Pell Grant program, but with fewer paperwork requirements and an award process that facilitates the identification of effects on college persistence, independent of initial college choice. We find that on average the grant increased neither enrollment nor credit attainment; the only notable positive average treatment effect was a 28 percent increase in the proportion of students completing 60 credits over two years but this was offset by a reduction in credits among other students. An exploratory analysis further suggests that the program's small average treatment effects mask considerable heterogeneity. In particular, it appears that students with a low (pre-randomization) propensity to persist in college received sizable positive benefits from the cash transfer, while students who were already more likely to persist in college received no benefit, and some may have been negatively affected. Thus, results from this experimental study suggest that the modest effects of conditional cash transfers for low-income university students primarily operate on credits rather than enrollment, and these could be enhanced with better targeting. More generally, we find that students respond to formal cash incentives in unexpected ways. Appended are: (1) Regression Coefficients for Propensity to Persist Equations; (2) Selected Characteristics of University Students at Baseline (Fall 2008) By Assignment to Treatment and Propensity to Persist Tercile; (3) Impacts on Educational Outcomes Over Three Years by Propensity to Persist: Model 1; (4) Heterogeneous Impact Bounds for First Stage Propensity to Persist Models; and (5) Quantile-Quantile Propensity to Persist Plots by Treatment Status and Model. (Contains 6 tables, 1 figure and 31 footnotes.