With the rapid development of text summarization, evaluation methods for automatic summarization system is becoming more and more important in natural language processing, which can promote development of text summarization greatly. This paper analyzes the existed methods for automatic summarization evaluation, and introduces a new evaluation method based on HowNet. The original tests have shown that the new method can touch the semantic meaning of the summarizations to some extent, thus reflect the quality metrics, for example, the sequence of sentences and conciseness much better. With more improvement in future, automatic evaluations will correlate with human assessments highly.