Guidelines are described for setting up an item bank under latent trait theory which may be applied to the achievement testing system of multi-section, large-enrollment, college survey courses. The enrollment for the course is typically heterogeneous: students may be majors or non-majors, any one section may contain honors college students and disadvantaged learners, and students may be of differing class levels. The advantage is that test scores can be standardized without legislating a common examination for all sections. Unique findings of this study were: (1) the computer file containing serial positions of items on examinations proved extremely useful; (2) obvious violations of unidimensionality and asking questions based on trivial, unevenly taught information led to items which did not fit the model well; (3) less stability in the equating procedure may result when large differences occur in an examination system; (4) more than one training session for instructors is recommended; and (5) latent trait models can work in a large, multi-section course examination which surveys a variety of ostensively different topics. (RL)