This paper looks at Fechner's law in the light of 150 years of subsequent study. In combination with the normal, equal variance, signal-detection model, Fechner's law provides a numerically accurate account of discriminations between two separate stimuli, essentially because the logarithmic transform delivers a model for Weber's law. But it cannot be taken to be a measure of internal sensation because an equally accurate account is provided by a χ2 model in which stimuli are scaled by their physical magnitude. The logarithmic transform of Fechner's law arises because, for the number of degrees of freedom typically required in the χ2 model, the logarithm of a χ2 variable is, to a good approximation, normal. This argument is set within a general theory of sensory discrimination.