OBJECTIVES: To review and synthesise the various methods used for cost-effectiveness (CE) modelling evaluating acute ischaemic stroke (AIS) treatment. METHODS: A literature search was performed in December 2006 through PubMed using the following search criteria: English language, publication date between 1995 and 2006, “cost-effectiveness” and “stroke” as key words. Only models were selected. They were reviewed according to a predefined reading grid. Structure, inputs and results were studied. RESULTS: Twelve models were retrieved, which evaluated thrombolytic agents or more general AIS management. The setting was either Europe, the US or Australia. The perspective was always societal. The general structure was similar between models (Markov model +/- decision tree) with a long-term horizon. States were defined by patients' vital status and disability level. However, some variability was found in the definition of disability (different scales, different thresholds) and in the choice of events to be included in the model (stroke recurrence, hemorrhage, cardiovascular event). Some inputs were comparable between models, namely probability of events and utilities per state. However, disability/mortality probabilities and costs per state were very different according to the country and calendar year. Long-term probabilities were often assumptions. Cost-effectiveness was most often expressed as cost per Quality-of-life Adjusted Life Years (QALY). One-way sensitivity analyses (SA) allowed identifying costs and utilities for dependent patients as drivers. Less often, probabilistic SA were performed in order to assess the robustness of the findings. CONCLUSION: The general structure of models evaluating CE of acute phase treatment in AIS seems to be consensual. However, there is a general lack of information on models' drivers. Influence of local AIS evolution data (mortality/disability) after acute phase and impact of disability on long-term outcomes should be tested. Once drivers identified, related robust epidemiological and economic data (notably long-term) should be used as input.