In this article, the statistical technique for setting up the consensus of perceived crime seriousness in previous studies is critically reviewed. The conventional method, when applied to a data set of crime seriousness, is found to have exaggerated the consensus because, by using a more appropriate model, which assumes perfect agreement in crime severity between subgroups, the consensus is reduced unanimously. By breaking down the population by gender, age, and educational level, sociodemographic differentials in crime seriousness are set up. The disparity is then further examined in details by paired comparisons between a target crime with fourteen other crimes. The three factors are all found to be important in affecting perceived crime seriousness. This conclusion is different from that of previous studies. The scaling method is found to be responsible for the difference. As the Thurstone method used in this study is more responsive and can produce more discriminating results, it is recommended for future research in crime severity. Finally, the implications of the findings on some important issues, like the appropriateness of legal penalty and the construction of a crime index, are discussed.