Recommendation engines have made great strides in understanding and implementing search personalization techniques to provide interesting and relevant documents to users. The latest research effort advances a new type of recommendation technique, Knowledge Based (KB) engines, that strive to understand the context of the user's current information need and then filter information accordingly. The KB engine proposed in this paper requires less effort from the user in representing the search task and is the first of its kind implemented in a digital library setting. The KB engine performance was compared with Content Based (CB) and Collaborative Filtering (CF) recommendation techniques and the text search engine Lucene by asking sixty subjects to perform two different tasks to find relevant documents in a database of 212,000 documents from 22 National Science Digital Library (NSDL) collections. Our KB engine design outperforms CB, CF, and text search techniques in nearly all areas of evaluation.