Empirical studies of the relationship between school inputs and testscores typically do not account for household responses to changesin school inputs. Evidence from India and Zambia shows that studenttest scores are higher when schools receive unanticipated grants, butthere is no impact of grants that are anticipated. We show that themost likely mechanism for this result is that households offset theirown spending in response to anticipated grants. Our results confirmthe importance of optimal household responses and suggest cautionwhen interpreting estimates of school inputs on learning outcomesas parameters of an education production function.