We use a bootstrap simulation framework to evaluate the relative importance of different sources of random and systematic error when estimating diet or food consumption of cetaceans, using a data set on harbour porpoise diet in Scottish (UK) waters from 1992–2003 (N=180) as a model. We also evaluate the consequences of applying explicit weightings to individual samples and/or sub-sets (‘strata’) of samples. In terms of the precision of estimates of diet composition, sampling error was the most important source of error, to the extent that overall 95% confidence limits changed only very slightly when sub-sampling error and regression errors were taken into account. On the other hand, for estimates of total food consumption by the porpoise population in Scottish waters, uncertainties about population size and energetic requirements were more important than uncertainty about diet composition. In relation to the accuracy of estimates of diet composition, the study also highlighted the importance of selecting regressions appropriate to prey in the study area (as opposed to ones constructed for the same prey species in another area) and demonstrated that applying equal weighting to individual samples or sample strata can substantially alter the resulting picture of diet. Therefore, the rationale for applying such weightings needs to be carefully considered.