Seismic facies analysis aims to characterize the seismic expression of a reservoir by placing seismic attributes into classes that are presumably related to geologic heterogeneities. Supervised seismic facies analysis can be used to classify the seismic expression of reservoirs in oil field development using well control as the training samples. Manual seismic facies analysis is often subjective, and it is usually time-consuming. Pattern recognition techniques such as support vector machine and neural networks have been used to automate seismic facies analysis. However, they often do not produce satisfactory spatial continuity in their results. The spurious spatial discontinuities are usually caused by the so-called “over-fitting” problem inherent in these methods. To avoid this issue, we have developed a supervised seismic facies analysis method based on image segmentation that promotes spatial continuity of the classified seismic facies. The seed region growing technique borrowed from image segmentation is used to segment attribute maps outward from points of well control. In addition, we calculate the seismic facies probability distribution at every grid point in the seismic image. We use this probability distribution as a weighting coefficient to allow regions to grow anisotropically. As demonstrated by the results on a field data set, this approach improves spatial continuity and indicates promising results.