Intensity resolution and loudness matching behavior for pure tones was studied in quiet and in two levels of noise (15 and 40 dB spectrum level) to test the predictions of several models for relating these phenomena. Five normally hearing listeners participated. The Weber functions in noise were elevated compared to the ones in quiet when comparisons were made at equal sound-pressure levels (SPLs). Loudness matching functions (dB SPL in quiet versus dB SPL in noise) showed a slope greater than unity. Modified power functions with threshold corrections were fitted to the matching data to estimate the loudness function exponent. The best-fitting loudness exponent for individual and group data (range = 0.24-0.35) was in the range of values typically found using magnitude estimation procedures for the loudness equation yielding the best fit. Three models for predicting Weber functions from loudness were evaluated with this loudness representation. One of these models, the subjective analog to Weber's law, yielded results inconsistent with the observed data. The other models, McGill and Goldberg's [J. Acoust. Soc. Am. 44, 576-581 (1968)] neural counting model and the proportional-jnd theory, predicted Weber functions that are consistent with the observed data if the variance of the decision variable is assumed to change in quiet and noise backgrounds. The proportional-jnd theory has such a change built into the model, but the underlying physiological mechanisms responsible for its success are unknown.