It has repeatedly been claimed by various authors that signal phase is more closely related to signal shape than is the amplitude spectrum. This is confirmed in an experimental study for the case of a visual evoked potential. It is shown that this may generally be expected in signals with a smooth amplitude spectrum. It is concluded that in such signals, quantitative analysis should therefore not be restricted to the amplitude spectrum. Some of the problems involved in phase analysis are also discussed and the application of tapered data windows in phase estimation is experimentally studied.