To　overcome　the　noise　in　personality　factors　and　precisely　predict　language　achievement,　we　propose　a　robust　regression　algorithm　based　on　the　maximum　correntropy　criterion　(MCC)　and　the　coarse-to-fine　method.　Firstly,　as　there　are　many　samples　while　few　personality　factors　correlate　to　the　language　achievement　in　the　data　set,　we　propose　a　regression　method　based　on　Pearson　feature　selection　to　eliminate　the　noise　and　redundant　features　for　solving　the　overfitting　problem.　Secondly,　as　the　learning　ability　of　each　traditional　regression　model　is　different　and　limited,　we　introduce　the　model　ensemble　method　based　on　MCC　to　predict　language　achievement　via　personality　factors.　Thirdly,　owing　to　the　fact　that　the　language　achievement　data　is　unevenly　distributed　and　the　same　model　parameter　cannot　fit　all　the　data　effectively,　we　propose　a　coarse-to-fine　prediction　method　to　reduce　prediction　errors,　which　divides　the　range　of　the　language　achievement　into　multiple　intervals　and　then　establishes　different　regression　models　at　each　interval　to　obtain　more　accurate　results.　The　experimental　results　on　the　data　set　of　the　personality　factors　and　English　achievement　demonstrate　the　high　precision　and　robustness　of　the　proposed　algorithm　compared　with　the　traditional　single　regression　models.