Concerned with the failure of multiple choice tests to measure composition--as opposed to grammar and usage--skills, college researchers studied the value of using writing samples to place students in composition classes. Evaluating the writing samples of 300 incoming freshmen, they did not find a high correlation between sample scores and subsequent achievement in courses. Believing that the lack of correlation possibly signaled evaluators' difficulties in maintaining a consistent performance, researchers replaced the more holistic marking of samples with grading based on an analytic, criterion referenced model. To further minimize problems inherent in large-scale writing assessment, they instituted a computer-assisted monitoring system that checked evaluators for agreement, adjusted student scores when divergence occurred in raters' responses, and provided placement recommendations based on the correlation of analytic raw scores and national test scores. Tracking 1,300 tested students as they moved through the freshman program, researchers found that the system minimized the chance for student misplacement, but, as all but 7% of the 565 students entering the basic composition course would have been placed there on the basis of their ACT or SAT score alone, they could not, as yet, give full endorsement to the system. (MM)