Measurement of training outcomes is a requirement for evaluating new training techniques, but is one that is different to meet. Managers of education and training may have different concepts of what they want, as favorable outcomes, than do the investigators doing the research. Classical statistical and experimental designs assume laboratory rigor of control over variables that is seldom possible in the real world of a school or classroom. Yet in the broader perspective of educational institutions, the effectiveness of the institutions is a current issue of fundamental concern in our society. In this report, possibilities for measuring outcomes of training are surveyed, considering training as a form of planned social change. Various approaches are discussed. Illustrations from the computer-assisted instruction (CAI) literature of recent attempts to measure training outcomes are given. The principal conclusions presented are that the classical four-way design is impracticable for most evaluation studies in training environments; that a policy of "adaptive research for big effects" is apt to be scientifically and administratively desirable; and that current attempts at measurement of training outcomes sill use fairly simple methods. (Author)