An experiment was performed to test the hypothesis that in a voluntary generalization task subjects encode the training stimulus (TS) in relation to the adaptation-level (AL) prevailing at the time of TS presentation. If so, then during a subsequent generalization test the point of maximal generalized responding should bear this same relationship to the AL of the generalization test series. To test this hypothesis, three major groups of subjects (n = 80) were given preexposures with dim lights (low preexperimental AL group), bright lights (high preexperimental AL group), or no preexposure (control group). Next, half of each group was exposed to a TS and a subsequent generalization (recognition) test including this brightness and four others. The other half of each group was exposed to the same TS and subsequent generalization stimuli, but their task was to rate the stimuli on a 7-point bright-dim scale. This provided AL measures which could be correlated with generalization performance in the comparable generalization groups. The preexposure condition was successful in producing the expected low or high preexperimental ALs, as reflected in different ratings of the TS when it was presented. Furthermore, these effects persisted throughout testing, as evidenced in ALs (i.e., rating scale midpoints) which reflected the preexposure condition. Finally, the difference' between the point of maximal generalized responding and the AL of the generalization test series was highly and significantly correlated with the difference between the original rating of the TS and scale midpoint. This supports the major experimental hypothesis of the study. Thomas and Jones (1962), in a study of voluntary stimulus generalization, found that the location of the training stimulus within the test series had a marked effect on the symmetry of the resulting generalization gradients. Their subjects were given a 6O-sec exposure to a 525-nm training stimulus (TS) with instructions to remember the stimulus so that they could identify it during a subsequent wavelength generalization test. The location of the TS in the test series was systematically varied, with the TS being in the center position of the test series for one group and asymmetrical (i.e., displaced from the center) for four other groups. Only with the symmetrical group did a symmetrical generalization gradient (with a peak at the TS) result. For the other groups, the peak was shifted from the TS toward the center of the test series. This finding, which Thomas and J ones called a "central tendency effect," has been replicated many times with different stimulus