As technology continues to advance in the semiconductor industry, the need has evolved for smaller and more complex images. The resulting tighter image size specifications have forced mask manufacturers to place increasing emphasis on process optimization to achieve control of image size variation. A key element of this optimization is the reduction of image size variation across a mask. As the within-mask variation specification tightens, it becomes increasingly important to identify all nonrandom sources of variability. These nonrandom, or systematic, sources of variation are those which are, theoretically, controllable and, therefore, candidates for elimination. If they can be identified as to type and amount of contribution, then the engineering community will have a better understanding of where to effectively focus its optimization efforts. This paper presents a simple yet powerful method of regression analysis for identifying both the type and amount of systematic sources of image size variation across a mask. The types of systematic variation discussed include radial, side-to-side, and axis (X-to- Y delta). Also presented are ways in which these sources of variation are identified and contributions quantified. Examples from simulation and actual product data are given.