Unidirectional elaboration of male trait evolution (e.g., larger, brighter males) has been predicted by receiver bias models of sexual selection and empirically tested in a number of different taxa. This study identifies a bidirectional pattern of male trait evolution and suggests that a sensory constraint is driving this divergence. In this system, the inherent trade-off in dichromatic visual detection places limits on the direction that sensory biases may take and thus provides a quantitative test of the sensory drive model. Here I show that sensory systems with trade-offs in detection abilities produce bidirectional biases and that signal design properties match these biases. I combine species-specific measurements and ancestral estimates with visual detection modeling to examine biases in sensory and signaling traits across five fish species occupying optically diverse habitats in the Californian kelp forest. Species-specific divergence in visual pigments correlates with changes in environment and produces different sensory biases--favoring luminance (brightness) detection for some species and chromatic (color) detection for others. Divergence in male signals (spectral reflectance of orange, blue, and silver color elements) is predicted by each species' sensory bias: color divergence favors chromatic detection for species with chromatically biased visual systems, whereas species with luminance sensory biases have signals favoring luminance detection. This quantitative example of coevolution of communication traits varying in a bidirectional pattern governed by the environment is the first demonstration of sensory trade-offs driving signal evolution.