Community-based question-answering (CQA) services contribute to solving many difficult questions we have. For each question in such services, one best answer can be designated, among all answers, often by the asker. However, many questions on typical CQA sites are left without a best answer even if when good candidates are available. In this paper, we attempt to address the problem of predicting if an answer may be selected as the best answer, based on learning from labeled data. The key tasks include designing features measuring important aspects of an answer and identifying the most importance features. Experiments with a Stack Overflow dataset show that the contextual information among the answers should be the most important factor to consider.