As domain-specific phenomenal terms are usually compounds that are difficult to extract according to local context features via the traditional machine learning methods,a novel extraction method is proposed. In this method,first,the context-based method is employed to extract the separator set. Then,with the combination of the separator set and context terms,the improved NC-value algorithm is used to extract candidate phenomenal results. Finally,nominal terms are filtered out from the candidate phenomenal terms to obtain the final terms. Experimental results indicate that the proposed extraction method of domain-specific phenomenal terms performs better than the word frequency-based and the separator-based ones.