We propose two image quality assessment metrics named multi-scale modular similarity (M2S) and multi-scale modular maxima similarity (M3S). It has been well known 1) multi-scale analysis is an effective decomposition technique in image processing, and 2) contours and edges analyses are crucial in the understanding of natural scenes. Motivated by these two facts, we attempt to develop quality assessment metrics using multi-scale edges presentation. We decompose an image with un-decimated dyadic wavelet transform, and then develop M2S metric to evaluate the quality of images by comparing the modulus across scales of wavelet transform. Multi-scale edges are defined as local maxima of modulus, which often contain the most important information of the image. As a further step of M2S metric, M3S only uses the multi-scale edge information. M3S is therefore essentially a reduced-reference image quality metric. Extensive experiments indicate that in most cases, the prediction abilities of these two proposed metrics are similarly excellent and both outperform the widely used PSNR and the simple structural similarity metrics.