Abstract: In high energy flash radiography, the blur is a key problem seriously affecting the accuracy of diagnosis. Due to blurring the detected boundary from the image has a deviation from the true value. The larger the blur size, the larger the boundary deviation with the radiographic layout and the structure of object fixed. On this basis, a new method for determining the blur size of the radiographic system from boundary deviation is presented in this paper. The principle of the method is explained from the physical process of flash radiography. The method is applied to the simulated images, and the measurement of the blur size is validated by means of radiographing the step object. It is shown that the FWHM error of the fuzzy function is less than one pixel, and the backward cone collimator used to photographing the spherical object is recommended.