To detect dim and small target in infrared image, a novel method based on cross window was presented through analyzing traditional methods. The method constructed cross detection window with up, down, left and right reference points around candidate detected point according to the size of target. Based on the grads gradient between candidate detected point and reference points, the points belong to target can be identified. The method does not need background prediction, and achieves higher detection probability and lower false alarm probability in infrared image with low signal noise ratio and strong background. Experimental results show the validity of the method.