OBJECTIVES: Estimates of mean cost and quality-adjusted-life-years are central to the cost-effectiveness analysis of health technologies.  They are often calculated from curve fits to overall survival and time on treatment, ideally by the method of maximum likelihood applied to individual patient data.  However, such data is often not available.  Instead, curves are commonly fit to summary Kaplan-Meier estimators, either by regression of the transformed estimator or by minimising the sums of squares of differences between actual and fitted values.  However, the tail of the estimator is often uncertain due to small numbers of patients at risk, and the curve fits do not yield estimates of the true uncertainty in survival times, which is a very important component of overall uncertainty in cost-effectiveness.  Here, I describe a new, more accurate method of fitting survival curves to summary survival data.  METHODS:        First, I estimate the underlying individual patient data from the Kaplan-Meier estimator, numbers of patients at risk and from other published trial-related information.  The fitted curve is then estimated by maximum likelihood given the estimated underlying individual patient data.  RESULTS:           Simulation applied to individual patient data shows that the method tends to give a more accurate curve fit than the traditional methods of fitting to the Kaplan-Meier estimator.  Furthermore, the curve fit is often very similar to that derived by fitting to the underlying individual patient data by maximum likelihood.  The method naturally yields accurate estimates of the uncertainty in survival times.  When applied to economic evaluations submitted to NICE, the method often yields substantially improved estimates of cost-effectiveness compared to estimates based on fitting survival curves in the traditional manner.  This highlights the sensitivity of many cost-effectiveness analyses to curve fits.  CONCLUSIONS:   When only summary survival data is available, I recommend the method for cost-effectiveness analysis.