Forty subjects participated in a study to test the accuracy of a recent model of human circadian phototransduction for predicting the relative effectiveness of two polychromatic light sources at suppressing nocturnal melatonin. Brief exposures to four different light levels (30, 100, 300 and 1000 photopic lux at the cornea) and two different "white" lamp spectra (4100 and 8000 K) were used. Results suggest that the model can properly order the relative magnitudes of the two circadian stimuli, but that nocturnal melatonin suppression follows a rate-limited response to light that cannot be predicted from the magnitude of the suppressing light stimulus alone. Some practical implications of these results are discussed.