Current Signal Detection Theory models of source memory necessitate assumptions about the underlying distributions of source strengths to describe source memory performance. The current experiments applied a modified version of the same-different task in order to plot individual memory stimuli along a controlled dimension of the average frequency of voices. This technique allowed us to determine that subjects were using an independent-observations strategy rather than a differencing strategy when deciding whether two test words were spoken by the same or different female speakers at study. By including two male and two female voices and changing the task distinction from same or different speakers to same or different genders, we predictably switched subjects’ decision strategies. With this new same-different memory design, we are one step closer to ending our reliance on measures that are inferred from data to describe subjects’ source memory performance.