In this paper, We propose a image watermarking method which combine frequence-domain with fuzzy inference. In our method, the original image is transformed and decomposed using DWT. The watermark is added to high-frequency coefficients, which analyzed optimally so as to genarate the fuzzfied data. In order to evaluate the robustness, the embeded watermark is detected in case of attacking by JPEG compression and cropping. Experimental results showed that proposed scheme is superior to the typical method with PSNR and similarity under the same conditions.