The ultimate goal of change agents working from below is to implement improvements that will affect coworkers, clients, and citizens. However, the close association of authority and decisionmaking responsibility creates two obstacles to change: red tape and autocratic supervision styles. Other obstacles include lack of rewards for change suggestions and resource limitations. These barriers are compounded by problems of the change agents themselves, such as errors made in developing change products, misconceptions about power, and lack of individual commitment to organizational effectiveness. In order to be able to initiate change from below, students of criminal justice must acquire the ability to apply scientific methods and to sell new ideas to personnel in positions of authority. The change agent training (CAT) program at the University of Maryland has been designed to teach students change agent skills. The program assumes basic knowledge in statistics, research methods, and organizational theory; students are also expected to have had some direct exposure to an operational agency. Two special sequential courses introduce students to use of scientific concepts and to selling ideas. In a further program component, students are responsible for managing evaluation projects which are being conducted for the parent agency. Students have two semesters to complete the evaluation projects. To achieve the training objectives, the program's teaching method seeks to aid the recognition, acquisition, and appropriate use of power. Students are involved in decisionmaking through a contract teaching approach. In addition, the classroom setting is unstructured, students play an active role in the learning process, and students are taught to deal with frustration. A supportive relationship is maintained with CAT participants for several years after graduation. Despite organizational barriers to change, students trained in this program should be able to effect improvements in the agencies in which they choose to work. The success of the program must still be empirically validated. Notes, tables, and a bibliography are supplied.