This paper investigates the experimental vibration technique of impact excitation when used to determine modal damping. Current practices introduce a consistent bias error in the derived modal damping estimates. The error can be significant, with levels of damping being wrongly predicted by a factor of three or more. This paper identifies the source of the error, and derives a simple correction to be applied to the observed modal damping estimates. The procedure is demonstrated by experiment and the results of a round robin exercise.