In this paper, we present a fall detection method for elderly people. The psychological burden, after the fall, and other environmental factors can result in serious harm when the elderly cannot get timely help. The main purpose of the fall detection system is to detect the fall of the elderly in a short time. The fall is characterized by high deformation of shape. Based on this later, we propose an approach based on the horizontal and vertical variations of human silhouette area. After human silhouette extraction, we extract the bounding box, then, we construct a histogram based on an intersection of vertical and horizontal partitioning for posture representation. Then, the fall detection is based on the combination of posture recognition and some specific rules. Experiments were conducted on different databases. Results demonstrate that the proposed method can detect fall effectively.