Smoking is among the most debated areas of risk perception, with varying conclusions about people's understanding of the risks. Part of the debate has focused on the measures being employed to support such claims. However, studies have not been conducted to compare the best methods to measure the perceived risk. This research first discusses the use of such measures, including survey data from tobacco industry archives that have not previously appeared in publication. Then, using nationally representative survey data of youth and adults in the USA, verbal probability scales and numeric scales are compared. The relationships between these measures are first examined for logical consistency with one another. Additionally, the strength of the relationships with and modeling power of the behavior of interest, smoking, are analyzed The results of difference of means tests, correlations and logistic regressions show that the numeric measures are inconsistent with logical semantic understanding, and more importantly, that vague quantifier scales show greater relationships and predictive power than numeric scales. Implications for survey design and further research are discussed.