Many developing country schools have four-hour school days and teach two groups of children each day. Governments are considering lengthening the school day, at great expense, to improve school quality. Advocates of the shift system argue the reform is unnecessary, as evidence from developed countries suggests increasing instructional time only improves achievement scores by small amounts. This paper is the first study of the e!ect of a large increase in instructional time in a low-income country. In 2005, the Ethiopian federal government directed school districts to abolish teaching in shifts and lengthen the school day from four to six hours. Districts implemented the reform at di!erent times, creating exogenous variation in instructional time. I use a di!erence-in-di!erence specification controlling for time-invariant unobservables at school level on a unique longitudinal dataset. For eight-year-old children, a longer school day improved writing and mathematics scores, but had no significant e!ect on reading. However, e!ects are larger among better-o! children: children who are not stunted, children from richer households and children in urban schools. The exception is that the reform has larger positive e!ects on girls than boys. The reform thus improves achievement on average, but may exacerbate gaps between wealthier and poorer children.