This paper presents a fast algorithm for computing shadow volumes of area light sources and performing shadow classification. The algorithm improves the Chin and Feiner's method in the way that it does not subdivide an area light source before the processing of scene polygons. Instead, a light source is only subdivided when needed and is merged immediately when the need vanishes. Much redundant work are saved by avoiding the unnecessary subdivisions. Comparing with the Chin and Feiner's method, experiments showed that the new method ran faster without loss of accuracy.