We investigate the reasons for there being an apparent limit to the accuracy of valuation using multiples, when it is implemented in a standard way. We find that most of the error comes from failing to correct known biases in earnings forecasts. Another part comes from mismatching observable...