At present, most social networking systems recommend potential friends mainly according to the existed friend relationship, and users' interests are not emphasized. Furthermore, it is a very difficult task to find users' interests with high precision from a large amount of data. A Friend Recommendation Based on user Tagging (FRBT) algorithm was proposed to find potential friends with the same interests by mining users' interests in tagging behavior data. First, Term Frequency-Inverse Document Frequency (TF-IDF) was used to cluster the similar semantic tags into topics. A new formula for calculating the users' similarity of topics was described. Combined with the user similarity based on topic and item, the proposed algorithm could recommend the users with high similarities as potential friends. The experimental results on tagging dataset of Delicious validate, compared wtih the algorithms of item, tag and tri-graph, FRBT has better performance in terms of precision and recall.