A variety of expensive software maintenance and testing tasks require a comparison of the behaviors of program versions. Program spectra have recently been proposed as a heuristic for use in pcrforrning such comparisons. To assess t,lrc potential 11sefulness of spectra in this cont,ext, we conducted an experiment that examined the relationship bct,wccn program spectra and program behavior, and empirically compared several types of spectra. This paper reports the results of that experiment,.