Subjects judged the length of lines, the heaviness of weights, or the roughness of sandpaper in 2 conditions. In 1 condition, they were instructed to make all their judgments relative to a long-term reference point, which consisted of a reference response and sensation. In the other condition, they were told to use a short-term reference point, namely, the response and sensation of the previous trial. A dynamic model of proportional judgment (L. T. DeCarlo, 1989/1990) predicts that the autocorrelation of successive responses will be larger for the latter instructions. This prediction was confirmed for the 3 continua. In addition, fits of a recently proposed dynamic regression model show that there is little or no effect of the previous stimulus intensity on the current response, whereas the results for an earlier model suggest a large contrast effect. The theory and experiments provide insight into judgmental and sensory processes in magnitude scaling.