This paper uses a large, nonexperimental data set to evaluate the effects of a preschool enrichment program in a developing country on cognitive, psycho social, and anthropometric outcomes. Outcomes are shown to be highly dependent on age and duration of exposure to the program. To minimize the impact of distributional assumptions, program impacts are estimated nonparametrically as a function of age and duration. A generalized version of the method of matching is developed and used to control for nonrandom selectivity into the program or into alternative program durations. The estimates obtained using this method reveal a different pattern of program impacts with respect to age and duration than does a parametric model under more restrictive functional form assumptions. The estimates based on matching show a greater dependence of test score impacts on duration of exposure to the program and show larger impacts for the anthropometric measures for a range of durations.  Impact estimates are based on three different comparison groups: children in the same communities in which the program was introduced who were not in the program, children in similar communities in which the program had not yet been introduced, and children who were in the program for a month or less. The average impact estimates for test score outcomes are robust to the three alternative comparison groups. The preschool program is found to increase cognitive and psycho social test scores, but only for children who participated in the program for at least seven months. The anthropometric results for weight differ substantially depending on which comparison group is used, suggesting that estimates based on the first two comparisons are contaminated by important unobserved characteristics related to program entry. The preferred estimates based on the third comparison group indicate that the program tends to improve the anthropometric outcomes, again with initially increasing effects as the duration of program participation increases. Cost benefit analysis based on these estimates and other assumptions indicate fairly high rates of return for this program.