OBJECTIVES Health technology assessment of treatments often requires estimates of their survival curves. Individual patient data (IPD) are often unavailable and the survival curves are usually calculated by fitting a nonlinear least squares (NLS) model directly to Kaplan Meier plots provided in the published literature. This method does not account for the uncertainty associated with the Kaplan Meier curve and can lead to biased estimates.  Although the IPD are often missing, the Kaplan Meier curve itself can be digitised and used to approximate what the original IPD could have been. METHODS We simulated trial IPD data from different survival distributions in order to assess the accuracy of the IPD reconstruction methods. The assessment of accuracy is made at multiple stages and ultimately the effects on the incremental cost effectiveness ratio (ICER) estimates are compared.  To do so, a simple cost-effectiveness model was developed, assuming two health states (alive and dead), and assigning costs (£1,000 per month plus drug costs) and a utility score (0.70) to generate ICERs.  Two additional methods to curve fitting are compared against the NLS approach – those suggested by Guyot (G), and by Hoyle & Henley (HH). RESULTS We find that the methods differ in accuracy at each of the following two stages; (a) model selection via the AIC and secondly (b) survival model parameter estimation.  When an underlying Weibull function was assumed, the ‘true’ ICER should be £28,924, compared against £31,182 £33,449 and £31,650 for the NLS, HH and G methods respectively.  When an underlying loglogistic function was assumed, the NLS, HH and GG methods produced ICERs of £26,507, £25,559 and £25,857, compared to a ‘true’ ICER of £25,779.  CONCLUSIONS These findings suggest that inherent biases may be apparent in each of the approaches, and these may manifest themselves differently, depending upon the ‘true’ shape of the underlying data.