Shot boundary detection (SBD) is the first step in video management and retrieval system. Existing SBD techniques, however, are mostly based on original frame histograms, which process too much data and are sensitive to flashes and small motions. Furthermore, they mainly perform boundary detection between each two frames, which are unreliable when there appear noises. To address these problems, we investigate a wavelet based shot boundary detection technique. It employs wavelet transform to get thumbnails of the original frames. Furthermore, it employs between cluster distances instead of adjacent distances to detect boundary and is robust to noises. Extensive experiments have been carried out on the platform of TRECVID benchmark. The experimental results show that the proposed technique outmatches some others and achieves satisfactory results.