The visual cortex of macaque monkeys has been divided into two functional streams that have been characterized in terms of sensory processing (color/form vs motion) and in terms of behavioral goals (object recognition vs spatial orientation). As a step toward unifying these two views of cortical processing, we compared the behavioral modulation of sensory signals across the two streams in monkeys trained to do a visual short-term memory task. We recorded from individual neurons in areas MT, MST, 7a, and V4 while monkeys performed a delayed match-to-sample task using direction of motion as the matching criterion. This task allowed us to determine if sensory responses were modulated by extraretinal signals related to the direction of the remembered sample. We sorted neuronal responses as a function of the remembered direction and calculated a modulation index, MI = (maximum response--minimum response)/(maximum response + minimum response). In the motion pathway, we found virtually no extraretinal signals in MT (average MI = 0.11 +/- 0.01 SE, 66 cells), but progressively stronger extraretinal signals in later stages, that is, MST (average MI = 0.17 +/- 0.01 SE, 57 cells) and 7a (average MI = 0.23 +/- 0.02 SE, 46 cells). In contrast to MT, strong extraretinal signals for direction matching were found in V4 (average MI = 0.28 +/- 0.02 SE, 94 cells), a relatively early stage of the color/form pathway, even though this pathway is not generally viewed as playing a major role in motion processing. Some cells in V4 were also tested while the animals performed a color matching task. These cells showed memory-related modulation of their response when either color or direction was used as the matching criterion. We conclude that extraretinal signals related to the match-to-sample task may be stronger in the temporal (color/form) pathway than in the parietal (motion) pathway, regardless of the stimulus dimension involved. Furthermore, our results indicate that the temporal pathway is capable of making a significant contribution to motion processing in tasks where motion can be considered as a cue for the identification of object attributes.