The brain processes external stimuli through special receptor cells and associated sensory circuits. In many sensory systems the population of neurons splits into ON and OFF cells, namely cells that signal an increase vs. a decrease of the sensory variable. This happens in brains from worm to man, and in the sensing of temperature, odor, light, and sound. For example, in the visual system ON cells respond to the onset of light, while OFF cells respond when the light is turned off. Here we analyze the possible benefits of this so-called “pathway splitting” using information theory. We find that the best way to divide a population of neurons into ON and OFF depends on the amount of neural noise, and what stimulus distributions the neurons try to encode. In the visual system, natural images contain many more dark contrasts than bright, therefore, our work predicts that there should be more OFF cells. This is indeed the case in the retina, the first order sensory organ that processes visual information. Our theory offers insight into this ubiquitous phenomenon of neural organization and suggests new experiments in diverse sensory systems, beyond the visual.