Subset size selection is an essential step for digital image correlation which affects the calculation accuracy greatly. The effects of pixels in a subset on calculation result are investigated. A new method is developed to deal with the problem. In the method, all the pixels in a subset are not treated equally as the traditional method. Coefficients are set to each pixel rely on the importance of them to identify a subset from target images which are calculated by the normal distribution based on the distances between them and the central pixel. Similar to subset size selection in traditional digital image, it is availabe to select appropriate number of pixels in subsets according to test conditions. The effect of pixels at the edge of subsets on correlation calculation is somewhat alleviated. Thus, the calculation accuracy is improved.