Mitochondria are important subcellular organelles that only appear in eukaryotic cells. The mitochondria proteins play important roles in energy metabolism. Knowing the protein localization inside the mitochondria is critical in understanding the protein function. Since experimental approaches for determining protein localization are always costly and time consuming, computational methods for predicting protein localization becomes an important topic. In this paper, we proposed a novel method, namely Weighted Gene Ontology Scores for predicting protein submitochondria locations. Our method performs better than the existing methods, indicating its potential application value in practical research.