In a laboratory experiment, human subjects were presented pairs of recorded 1-min traffic sounds. Subjects chose whichever sound of each pair they thought they would rather be exposed to on a regular basis; they also judged the relative dissimilarity of the sounds in each pair. The data were analyzed using multidimensional scaling techniques. The choice data showed that relative aversiveness of the sounds was related mainly to their subjective intensity. The dissimilarity data showed that subjects distinguished sounds on the basis of both subjective intensity and information, independent of intensity, about the source of the sounds. The single physical noise measure most highly correlated with the subjective intensity attribute from both sets of psychological data was Leq, the energy-equivalent sound level. Several new measures of the shape of the average power spectrum, and of the time-dependent variability in dBA levels, failed to add appreciably to the predictive power of Leq alone. Subject Classification: [43]65.75, [43]65.50; [43]50.70, [43]50.50.