When different health care interventions are not expected to produce the same outcomes both the costs and the consequences of the options need to be assessed. This can be done by cost-effectiveness analysis, whereby the costs are compared with outcomes measured in natural units--for example, per life saved, per life year gained, and per pain or symptom free day. Many cost-effective analyses rely on existing published studies for effectiveness data as it is often too costly or time consuming to collect data on cost and effectiveness during a clinical trial. Where there is uncertainty about the costs and effectiveness of procedures sensitivity analysis can be used, which examines the sensitivity of the results to alternative assumptions about key variables. In this article Ray Robinson describes these methods of analysis and discusses possibilities for how the benefits of alternative interventions should be valued.