When printed materials are used to commit crimes, such as threatening letters, font identification is used to estimate the equipment that was used in their production. To identify the font of printed materials, comparison testing with a print reference sample is required. As retrieving the target print sample from a huge volume of samples requires significant time and effort, a method for automatically searching the font is required. In this paper, we propose a similarity measure using SIFT features, which are invariant local features to the zoom and rotation of the character, and a method of retrieving fonts using the similarity measure. Retrieval experiments were performed using 102 font types with 2,230 types of characters for each. The proposed method is effective for retrieving fonts, even when the character size of the query image and the database is different. Furthermore, we propose a method of retrieving fonts at high speed through dimension reduction using principal component analysis. We show that through the proposed speeding up method, the retrieval time can be reduced to one-quarter of the previous speed without lowering retrieval performance.