The hypothesis that people can make non-learned numerical estimates of ratios of subjective magnitude is still unverified after over more than a century of research. Participants were asked to numerically estimate brightness ratios of pairs of surfaces whose luminances were combined factorially. With one luminance in a pair as the parameter, the ability to estimate ratios numerically predicts that the plotting of numerical estimates against the other luminance yields diverging curves that increase in height with the range of luminances. The present study found that statistically such curves were parallel and their height was constant with luminance range. These results support the hypothesis that people numerically rate subjective differences while they are performing the task to numerically estimate subjective ratios. Implications for theory and practice are discussed.