We show here evidence for the necessity of an asymmetric form in modelling the profile of an acoustic mode in the power spectral density. The analysis was performed on a 805-day series of GOLF data (l=0 to 3). The assumptions used for the fits are discussed and their consequences quantified, in particular for the optimum choice of the fitting spectral window. Values are given for the bias on the mode parameters (frequency, width, splitting) when using a Lorentzian fit. The bias depends on the degree l and on the frequency, and when taken into account leads to variations in the mode parameters with degree more consistent with theoretical expectations.