The direction of the motion which blurs the image can be dealt with as unchanging during the short exposing time. If it is identified, the blurred direction can be rotated to the horizontal axis, and the image restoration can be worked out easily in one dimension. An excellent simple model for imagery statistics is that of a spatially isotropic first-order Markov process. The autocorrelation of the original image and its power spectrum are assumed to be approximately isotropic. The motion blurring decreases the original image's high frequency contents in the motion direction. Thus, a derivative of the image in the motion direction would suppress more image intensity than a derivation in other direction. Then the motion direction is identified from the blurred image. The derivation matrix is the key for the identification. We select a propriety unchanging step for the direction derivation, a 3×3 direction derivation matrix is then constructed by using the double line interposition. This 3×3 direction derivation matrix can help to identify any motion directions of the most motion blurred images with high precision. It is steady-going.