The paper provides a quantitative basis for receiver design whereby timing performance can be evaluated for an arbitrary impulse response using known or measurable system parameters. The analysis accounts for the statistical independence of the multiplication processes of the dynodes as well as the possibility of their non-Poisson behavior. Fixed threshold, constant fraction, and median receivers have been analyzed, and the median receiver has been shown to be significantly superior. Experimental results have been presented which show reasonable agreement with the theory and which confirm the superiority of median timing.