Recent advances in music information retrieval have enabled users to query a database by singing or humming into a microphone. The queries are often inaccurate versions of the original songs due to singing errors and errors introduced in the music transcription process. In this paper, we present the Time-Warped Longest Common Sub-sequence algorithm (T-WLCS), which deals with singing errors involving rhythmic distortions. The algorithm is employed on song retrieval tasks, where its performance is compared to the longest common subsequence algorithm.