Learning characteristics of students undertaking computer assisted instruction using the PLATO system were assessed for five Russian courses at the University of Illinois. Since the computer allows accurate measurement of the interaction of the student with the material being learned, four measures of student performance were assessed--error rate, rate of interaction with the material, total time, and review time. These measures were compared to each other as well as to semester grades in Russian courses. For five courses, the error rate during the learning process was significantly and negatively correlated with final grades (i.e., students with a high error rate had low grades and vice versa). Final grades were based on instructor produced and administered measures rather than on computer materials. For each course, the highest grades were received by students with a low error rate and high interactions per minute (i.e., the number of times in each minute that the student requests a response from the computer). For all five courses, students with a low error rate and high interaction had a greater percentage of review time than did students with a low error rate and low interaction. In other words, more review led to a better grade. The total time spent at the computer terminal was not a significant indicator. Some students put in a great deal of total time and review time and still received the lowest grades. The data support a model for predicting performance of superior language students (those who learn in a fashion which minimizes errors) but show a need for further investigation of the performance characteristics of weaker language students. Four groups of students were distinguished based on error rate and interaction rate. (SW)