We conduct a randomized evaluation of a school library program on children's language skills. We find that the program had little impact on students' scores on a language test administered 16 months after implementation. The estimates are sufficiently precise to rule out effects larger than 0.13 and 0.11 standard deviations based on the 95 and 90 percent confidence intervals. The finding of zero effects is robust to different modes of implementation, individual tested language competencies and various subsets of the student population. We also find no impact on test scores in other subjects or on school attendance rates.