We developed a program for mapping local radiology system terms to LOINC that returns a ranked list of candidate LOINC codes based on document similarity scores. We compared the performance of this program with the Intelligent Mapper (IM) program in mapping diagnostic radiology terms to LOINC. The cosine similarity score ranked the correct LOINC code first in 34% of the terms in our development set and 39% of the terms from our test set, compared with IM's ranking of the correct LOINC code first in 83% of the terms in our development set and 92% of the terms in our test set. This study demonstrates the challenges in using document similarity scores for mapping to LOINC. Because vocabulary mapping is a resource-intensive step in integrating data from disparate systems, we need continued refinement of automated tools to help reduce the effort required.