Previous researches on query classification have focused on classifying web queries into categories of topic, search intent, geo-intent, product, and con-cept. However, no studies have ever attempted to find user goal on query in detailed level. In this study, we use a representation form of a fine-grained user goal (e.g., ‘verb + object’) and identify the fine-grained user goals of query from large scare how-to knowledge. In addition, we propose a meth-od for discovering the fine-grained user goals of the query by using search technique and supervised machine learning with retrieval related and linguistic features. We employed snippet of corresponding clicked URL of query, which has evidences in finding fine-grained user goals. In the exper-iments, we demonstrate the high performance in identifying fine-grained user goals of query and which features are effective. Finally, we present several findings lessoned from this study.