Shot transition detection is a basic step in video content analysis and retrieval. A number of automatic detection techniques exist but they leak their performance in the detection of gradual transitions. These techniques often suffer from high false detection rates due to the presence of illumination, camera or object movements in the video sequences. In order to reduce the influences of these effects in the detection process and to detect long dissolve and fade regions in the video sequences, we propose a modified color layout descriptor with a new similarity metric; suits well for the extracted features to identify and classify the gradual transitions. Experiments are carried out on various video sequences taken from TRECVid dataset and publically available dataset. The results show that the proposed method yields better result compared to that of the existing gradual transition detection methods. © 2012 Springer-Verlag.