BACKGROUND: Extrapolation is often a key element in health economic modelling. Although any model should use empirical data if possible, the effects of treatments on long-term health outcomes are seldom observed within the follow-up time of a clinical study. Extrapolation over a lifetime horizon will generally be required in economic models where treatments have different cumulative survival at the end of the clinical trial. Typically a within-trial analysis of costs and health effects, in which outcomes are truncated at the conclusion of the trial, will be overly conservative. OBJECTIVES: The purpose of this study is to compare different methods of extrapolation in the context of examples concerning oncology, although the principles apply across all therapeutic areas.  METHODS: There is a set of standard assumptions regarding extrapolation of survival data from clinical studies, ranging from very cautious (‘stop-and drop”) to very optimistic (“continued benefit”). The impact of different assumptions regarding extrapolation is explored, and the implications are discussed. CONCLUSIONS:  The choice of extrapolation method has significant impact on comparative clinical effects, costs and cost effectiveness. Based on our findings and supporting examples, we propose the following: 1. Analysts should perform and report results under a range of specific standard extrapolation assumptions to increase comparability across studies.  2. The choice of a base-case approach in any particular study should be guided by knowledge about the biology of the indication under evaluation and the mechanism of action of the treatment.  A case could be made for a reference case method of extrapolation, but we believe that sensitivity analysis across a standard set of possibilities is sufficient.  Adherence to these modelling practices will contribute to increased transparency in modelling and hence potentially to a greater confidence among health care decision makers in the results from cost-effectiveness analyses building on modelling and extrapolation.