Hydrocarbon exploration is geared towards identifying complex stratigraphic features in seismic data. Channel facies, of the most common stratigraphic features, are important from the viewpoint of hydrocarbon exploration. Depending on burial depth and fluid contents, they may be considered as potential reservoirs or drilling hazards. In this research, the mean-shift clustering algorithm has been used to improve the quality of channel boundaries in seismic data. Mean-shift is a non-parametric process not dependent on initial knowledge about clustering parameters. A Sobel operator is applied to the mean-shift clustering of seismic data. Finally, the results of the proposed workflow are compared with those of the similarity attribute, as a conventional attribute in channel detection. The combination of the mean-shift clustering algorithm with the Sobel operator eliminates many of the false edges and detects channel boundaries with a high resolution.