For image denoising by wavelet shrinkage, an optimal threshold selection is required. Although Donoho's universal threshold gives the optimal threshold theoretically, the information of noise power on the image is needed. In this study, we propose a simple and robust method for the threshold selection. We use the histogram shape of absolute value of wavelet coefficients. Medical chest x-ray digital image of various signal-to-noise ratios (SNR) were chosen for testing our threshold selection. For the noised image of a low SNR, the almost same performance as the Donoho's threshold was shown. Furthermore, the threshold value which we proposed was stable over the wide range SNR.