Pattern width homogeneity in an exposure field of gradient-index (GRIN) lens array was greatly improved by changing the sub-scan method for averaging the pattern widths. By investigating the pattern width distribution in the exposure field, it was clarified that the parts, where the printed patterns were degraded and pattern widths were notably changed, appeared as striped lines. It was supposed that the striped abnormal pattern-width change was caused by the extra exposures for the times when the sub-scan stage was stopped at both the sub-scan ends for turning the scan direction. For this reason, sub-scan length was vastly extended, and the sub-scan stage was turned at the both ends after all the patterns on a reticle passed over the actually used parts of a GRIN lens array. By this method, no patterns were exposed and projected on a wafer while the sub-scan stage was stopped for turning at the both ends. As a result, 15-µm L&S patterns were almost homogeneously printed in the exposure field within a variation range of ±6%.