Considering that the challenges in using peer grading to select the best-k students in MOOCs (massive open online courses) are twofold: first is strategyproof, i.e., students should not benefit by untruthfully reporting valuations; second, instead of exerting (costly) efforts to evaluate, students may randomly provide (or just guess) the evaluations on other peers. This paper proposes a truthful grading-based strategyproof peer selection scheme for MOOCs, TSP. Specifically, all students are partitioned into same-size clusters, and each student only evaluates students in other clusters. Moreover, peer prediction mechanism was utilized to motivate each student to truthfully report their gradings through comparing their reports with other peer students who conduct same evaluation tasks. The theoretical analysis and simulation results show that TSP can both stimulate students to truthfully reporting their gradings, and meanwhile select best k-students in strategyproof way.