OBJECTIVES Overall survival (OS) data from clinical trials in oncology are often incomplete, thus modelling over the lifetime horizon requires long term extrapolation and it is a critical input to cost-effectiveness studies. Data from the Surveillance, Epidemiology, and End Results (SEER) program may provide good validation on the long term OS. The objective was to examine the parametric functions that best fit data in lung cancer (LC) of various histologies in SEER. METHODS SEER data (2004-2011) were analyzed for patients diagnosed with stage IV small cell, large cell, squamous cell carcinoma and adenocarcinoma of the lung with complete follow-up. Mean age was 68.03 (sd 11.67) and 55.5% were males, with varying baseline age and gender distribution by histology.  Treatment status could not be established. Parametric models for OS were fitted using exponential, Gompertz, loglogistic, lognormal, and Weibull distributions. Models were fitted with and without covariates. Fits were inspected and compared graphically using survival and quantile-quantile plots, and statistically using the Akaike Information Criterion (AIC). Modelled mean life expectancy results were compared to the restricted mean life expectancy of the Kaplan-Meier estimator. RESULTS The lognormal distribution was found to have the best fit within the SEER population, both with and without covariates indicating that a small proportion of patients survive for a long time despite the poor general prognosis of any type of LC. Loglogistic and gamma distributions were 2^nd and 3^rd best, followed by Weibull, Gompertz and exponential, for all histologies. The last three fitted the data poorly, and underestimated mean life expectancy. CONCLUSIONS Only small proportions of LC patients are alive at 5-8 years, nevertheless the mean OS estimates are impacted by the choice of survival function. The lognormal distribution fit best across all histologies indicating a higher propotion of patients alive than estimated with Weibull models.