A blurriness metric method based on Maximal Oriented Gradient (MOG) is proposed. In this method, a set of filters are used to get oriented gradient maps from original image at first. Then, oriented masking is used to suppress gradient values. At last, a set of MOGs, which is extracted from the suppressed values, are used to predict the blurriness of the original image. Experiments of algorithm consistency and sensitivity show that new method has a good performance, and experiments on cross-distortion data prove that the proposed method is stable and applicable.