A portion of the variation between the observers is due to real differences in the contributions of the cone types to the different visual mechanisms, and another portion is due to experimental error. To investigate the effect of small variations in the real contributions of the L, M and S cones on the wavelength-responsivity functions, we performed the following simulation. First, we estimated the mean color-responsivity function, averaged across the four observers, for each of the three mechanisms. Then, we added a normally distributed noise term to each of the weights. The noise had zero mean and SD = 0.1 for the cone contributions to the black/white and yellow/blue mechanisms, and SD = 0.05 for the contributions to the red/green mechanism. When we used these SDs for the simulation, the variation we observed in the simulation was similar to the variation we observed in the functions estimated from the four subjects. Fig. 11(B) shows the results of sampling ten color-responsivity functions created by this random process. The variation that we observed in the simulation is similar to the variation we observed in the functions estimated from the four subjects.