A new method based on the linear model of image gray level was proposed to estimate sub-pixel jitters of a sequence image for cloud scene acquired by staring remote sensing imaging systems. Firstly, the image pixel and corresponding neighborhood gray levels were mathematically described by using linear model with three parameters, and the image gray was modeled. Then, by taking the jitter parameter as optimization variables in the linear mathematic model and the comparability between image sequences and reference frame as optimization objective, a new estimation method for the sub-pixel jitter was proposed based on least square optimization approach. Subsequently, the solving equation was derived. Finally, the method was verified by using simulated image sequences containing the cloud scene. Experimental results indicate that the proposed method is able to implement the subpixel estimation effectively and offers the estimation accuracy no less than 0.1 pixel. The obtained estimation accuracy is higher than that of the conventional feature-based ones, and can be used in geometric calibration and target positioning of remote sensing images as well the multi-frame relative detection of time-series images.