The Sexual Opinion Survey (SOS) has been one of the most widely used and influential measures in sexuality research cross-culturally since the 1980s. This scale is designed to measure individuals’ tendencies to avoid or approach sexual stimuli, also known as erotophobia-erotophilia. Although evidence of the scale’s validity has accumulated over the years, much of this work is limited to investigations of convergent and divergent validity. It has also been decades since the criterion validity of the original items has been reviewed and in that time some conceptual gaps in the scale’s item coverage have become apparent. Specifically, in this research we sought to determine if either the original or new SOS items of our own devising were better able to predict affective responses to a large range of sexual stimuli, compared to a variety of other sexuality measures (e.g., sociosexuality, sexual compulsivity, sexual health, etc.). To this end, we employed a machine learning method called Random Forests, in which all questionnaire items were entered as separate predictors. Results of this analysis indicated that a combination of the original and new SOS items better predicted affective responses to sexual stimuli than other sexuality measures that were tested. These findings confirm the validity of some of the existing SOS items but also suggest that further measurement refinement may be needed. Moreover, results of the exploratory factor analysis on the new SOS items identified a five-factor solution. Future validation efforts should confirm this structure.