Specifies the design for a national evaluation of the Head Start program but recommends a system of careful studies in preference to a national evaluation. Discusses normative and theoretical problems of defining the criterion variable, "social competence." Longitudinal evaluation of Head Start is not recommended because of inadequacies in child development theory, measurement incomparabilities, sample mortality, and inappropriate expectations for effects of limited social intervention. Evaluating a major social program should take account of the multiple constituencies by allowing local constituents to choose a small number of outcomes and to assign their own priorities to cross-site outcomes. Head Start places important constraints on the technical design; i.e. treatment variations, site, and children characteristics are confounded. Children should be randomly assigned within-site to treatment and control conditions; a less-preferred alternative is one based on growth curves (value-added design). Bases for choosing specific statistical analyses of evaluation data are recommended. A final recommendation is to pilot test the total evaluation. 450 pp. Bibliog.