In the process of bridge crack detection based on intelligent image processing, the number of images collected is huge, and the detection is very time-consuming. So, a fast screening method for bridge crack images is proposed. First, the edges in the image are extracted and the longest edge is found. Secondly, the minimum circumscribed circle related to the longest edge is obtained and then the circle radius is used to filter and classify the image by comparing with a threshold. The threshold is automatically determined by building an adaptive calculation model based on image resolution. Experiments show that the proposed method can quickly and accurately screen the crack images from the non-crack images, which greatly improves performance of the crack detection system.