A number of experiments were conducted to investigate the interaction of the ON and OFF pathways in the processing of global-motion signals. The stimulus employed was a variant of that used by Newsome and Pare [(1988) Journal of Neuroscience, 8, 2201-2211] in which a small subset of dots move in a common (global-motion) direction in a field of randomly moving dots. The threshold measure was the number of dots required to move in the global-motion direction for that direction to be detected. We found that: (1) the extraction of a global-motion signal carried by light dots (luminance above the background) was impaired by the addition of dark dots (luminance below the background) which did not carry the signal (noise dots); (2) sub-threshold summation occurs for global-motion signals carried by light and dark dots; and (3) a signal dot which changed luminance polarity (went from light to dark) did not result in a motion signal--either in the global-motion direction or in the opposite direction (reverse apparent motion). From these findings we conclude that the inputs to the motion sensitive cells have matched spatial opponency (the ON and OFF pathways remain separate at this level) but that they then combine to form a single pathway prior to the extraction of the global-motion signal. These findings are contrary to those predicted by models which advocate squaring or full-wave rectification prior to global motion processing.