Judgments of similarity, and thus the ways in which stimuli are grouped together to form categories, depend in part on the stimulus features, or dimensions, to which perceivers attend. Previous studies have demonstrated that emotional state biases attention toward emotional dimensions of stimuli. However, the studies do not establish that those dimensions are actually used in judgment, or quantify the weight given to the attended dimensions. The present experiments used a multi-dimensional scaling analysis to assess the weights that happy, sad, and neutral-emotion participants give to emotional and non-emotional dimensions of face stimuli in judgments of similarity. It was predicted that emotional participants would overweight the emotional dimension of the faces (i.e., emotional expression), and underweight other face dimensions, relative to neutral-emotion participants. This effect was observed in three experiments, both when the emotional dimension was salient and when it was relatively obscure, and using two different inductions of emotion. Evidence for emotion-congruent dimension use was also observed in one experiment. Results are discussed with respect to emotional response-based categorization (Niedenthal & Halberstadt, in press), the tendency for stimuli to cohere as categories on the basis of the emotional response the elicit in the perceiver.