As Introductory Programming classes continue to grow, it becomes ever more important to ensure that automatic grading is both efficient yet still maintains high accuracy. Specifically, the use of automated testing for this type of grading has become standard, but it is not without fault. Instructor test suites can often misclassify correct student submissions as incorrect or vice versa. To address this discrepancy, we propose a novel approach which adapts the technique of Test Suite Augmentation to an educational setting in order to improve instructor test suite quality. Our approach first partitions student programs into clusters based on Behavioral Equivalence. We then use these clusters to better guide the Pex test-generator in generating a minimal set of tests which can augment an instructor’s suite and improve grading accuracy. We evaluated our approach on an extensive set of student submissions from a CS1 course and were able to find flaws in existing instructor test suites, and then augment these suites to substantially improve their quality.