Manual retrieval of key frames from an Intravascular Ultrasound (IVUS) pullback sequence is a tedious and time-consuming task. The retrieval results are subjective and highly depend on clinical experiences and professional knowledge of the operator. In this paper, two methods for automatically retrieving key frames in an IVUS image sequence are proposed. One is based on the description of the vascular morphology. First, the morphological profiles of vessel are extracted. Then, the sequences of cross-sectional areas in each frame are coded with the symbolic aggregate approximation (SAX) algorithm. Finally, key frames are selected through comparing the Euclidean distances between the coded area sequences of successive frames with a properly selected threshold. The other is based on the histograms of IVUS images. The key frames are retrieved through comparing the Bhattacharyya distances between histograms of successive frames with a properly selected threshold. The validity of both methods has been validated in vivo image data and the accuracy was quantitatively evaluated. The results show that a higher compression ratio, lower redundancy and higher precision can be obtained with the morphological description method than the histogram method. However, its efficiency is relatively lower than the histogram method.