Donor agencies, and recipient governments want to assess the effectiveness of aid-supported sector policies. Unfortunately, existing methods for impact evaluation are designed for the evaluation of homogeneous interventions ("projects") where those with, and without "treatment" can be compared. The lack of a methodology for evaluations of sector-wide programs is a serious constraint in the debate on aid effectiveness. We propose a method of statistical impact evaluation in situations with heterogeneous interventions, an extension of the double differencing method often used in project evaluations. We illustrate its feasibility with an example from the education sector in Zambia. © 2008 Elsevier Ltd. All rights reserved.