A preceding vehicle detection method for forward collision warning system is proposed. The proposed method utilizes the horizontal and vertical edges instead of underneath shadow to detect preceding vehicles without being influenced by variant weather conditions. We first extract horizontal edges and then check the vertical edges above horizontal edges to confirm the location of a horizontal edge just beside an object and generate a candidate vehicle with vertical borders while horizontal edge is located at the bottom of the object. If the bottom of the object can't be found by a horizontal edge, we find the vertical borders and bottom of the object by searching symmetric vertical edge pair. Then, we estimate the width of the object to select candidate vehicles. At last, we use SVM to verify the candidate vehicles. In experiments, the proposed method was evaluated on variant weather conditions such as sunny day, cloudy day, and rainy day. The detection rate of preceding detection is over 90% in sunny day and cloudy day and is over 80% in rainy day.