Automatic document analysis tools for mathematical texts are necessary to enlarge the pool of mathematical knowledge available in electronic form. However, development of such tools is currently hindered by the weakness of optical character recognition systems in dealing with the large range of mathematical symbols and the often subtle but important distinctions in font usage in mathematical texts. Research on developing better systems for mathematical optical character recognition crucially depends on having an extensive, high quality database of glyphs used in mathematical texts for training and test purposes. We present such a database of symbols constructed from a large set of characters available in the LATEX document preparation system that can serve as a basis mathematical text recognition. We describe its integration into a prototypical system optical character recognition system for mathematics that enables the construction of LATEX source documents from mathematical documents available as images. From the lessons learned in this work we derive a road map for further research into the area of mathematical text analysis.