A computer implemented method and system for selecting a string for serving as a reference string for a comparison scheme for compressing a set of strings calculates preliminary compression results for every string relative to an initial reference string, and uses the preliminary compression results to find a better reference string without additional compression tests. According to one embodiment, a histogram is calculated showing the number of occurrences of each compressed length for each string in the set plotted against the initial reference string and the better reference string has a length corresponding to an average compression length or center of gravity of the histogram.