Evaluation of ongoing educational programs must necessarily differ from the basic research design; it must change to meet the changes of the program and its environment. Over the three years of the operation of the Philadelphia "Let's Be Amigos" bilingual program, the kinds of data generated in the program evaluation have evolved in response to the demands of project management, community and intra-school-system relations and the Office of Education. The evaluation of process aspects and product aspects of the program have evolved in opposite directions: (1) evaluation of the pupil performance program outcomes has tended to evolve from informal, criterion-referent approaches to more rigorous experimental designs; and (2) evaluation of processes has tended to evolve from formal methods (observational checklists, forced-choice questionnaires) to less rigorous methods (open-ended questionnaires, interviews, etc.). In the first operational years, assessment of pupils' reading was primarily criterion-referent, involving a word-calling test. The assessment of reading skills was modified after first-year evaluation, first passing through a phase in which an attempt was made to prepare materials-derived, criterion-referent tests to assess more complex skills, and from there to standardized tests. Evaluation of curriculum development has evolved from use of a formal checklist to use of an interview structure with open-ended questions. (KM)