Binocular processing was investigated using a quantitative, process-oriented meta-theory of respnse times. The analyses are not confined to particular distributional assumptions or specific models. Upper and lower performance boundaries for probability summation in parallel processing are defined and compared with observed distributions of reaction times using a variety of dichoptic stimuli. Performance that exceeds the upper bound strongly suggests facilitatory convergence between the two eyes (binocular channel summation). Performance below the lower bound suggest that inputs to each eye are processed serially. The results indicate that binocular channel summation in subjects with normal stereo vision requires targets of the same luminance polarity (paired increments or decrements) in corresponding retinal locations. When corresponding retinal locations are stimulated with opposing luminance polarities (increment to one eye, decrement to the other), performance is consistent with probability summation, indicating parallel ON and OFF pathways remain segregated at least to the level of binocular fusion. Further analyses of data from a stereoblind observer suggest serial processing of binocular inputs.