OBJECTIVES: Although there is much experience with using clinical evidence for healthcare policy decisions and patient preference research is becoming more prominent, there is limited experience in formally integrating the two types of evidence. The aim of this study is to show using two case studies how patient preferences can be used to weigh clinical evidence in a probabilistic multi-criteria framework, and to identify remaining methodological challenges. METHODS: The first case study compares three hypothetical antidepressants and placebo for the treatment of severe depression on four criteria. The second case study compares eight highly active antiretroviral therapies (HAART) for HIV-positive persons on four criteria. Preferences from patients and/or clinicians were derived from previous preference studies and clinical evidence was obtained from clinical trials. Univariate and multivariate probability distributions for the preferences and clinical evidence were combined using Monte Carlo simulation methods. The main model outcomes were treatment value distributions. Decision uncertainty was estimated with the probability of rank reversals for the first rank. RESULTS: In the antidepressants case, there seemed to be more decision uncertainty for clinicians (49%) than for patients (27%), and the decision uncertainty depended more on uncertainty in the clinical evidence (difference 23%). The decision uncertainty among patients in the HAART case was higher (64%) and depended slightly more on uncertainty in preferences (difference 3%). CONCLUSIONS: This study shows how elicited patient preferences can be formally used to weigh clinical evidence in a framework that explicitly considers uncertainty. The model can help increase insight into the decision and point to critical uncertainties in the evidence. Further work is required on integrating heterogeneity in preferences and clinical evidence, on quantifying decision uncertainty with value of information metrics, and on homogenizing evidence collection methods for the use in integration models.