This paper reports the conclusions of a recent workshop that was established to discuss how health impact assessments (HIAs) might be evaluated. The main purposes of HIA are: (a) to predict the consequences of different decisions; (b) to make the decision-making process more open by involving stakeholders; and (c) to inform the decision makers. 'Prediction', 'participation' and 'informing decision makers' are thus the three domains in which HIA should be evaluated. In the 'prediction' domain, process criteria scrutinize the methods used to see if it is likely that they would produce reliable predictions. Outcome criteria involve verifying the predictions, but this is frequently impractical and predictions for the counter factual (the option not chosen) can never be verified. In the 'participation' domain, process criteria examine the ways in which stakeholders were involved, while outcome criteria explore the degree to which the stakeholders felt included. In the 'informing decision makers' domain, process criteria are concerned with the communication between decision makers and those doing the HIA, and should reflect upon the relevance of the HIA content to the decision makers' agenda. Outcome criteria explore the degree to which the decision makers considered that they had been informed by the HIA. This paper concludes with suggestions for the types of information that should be included in HIA reports in order to permit the readers to make an assessment of the 'quality' of the HIA using the three domain criteria outlined above.