OBJECTIVES: Network meta-analysis of published survival data are often based on the reported hazard ratio. In this paper we illustrate the value of reconstructing data from published Kaplan-Meier curves to perform network meta-analysis. METHODS: Published Kaplan-Meier survival curves of trials evaluating different interventions for non-small-cell lung cancer were digitally scanned. Next, a dataset was created with for each treatment of each trial the number of events and number of patients at risk for multiple short time intervals over the complete follow-up period. Two types of network meta-analyses were performed: 1) For each publication for which no hazard ratio was reported, a hazard ratio was estimated based on the scanned curves. Subsequently, all hazard ratios of all trials were synthesized with a network meta-analysis model assuming a constant hazard ratio (2-step approach); 2) A network meta-analysis of the constructed data of the Kaplan-Meier curves of all trials was performed (1-step approach). RESULTS: The 1-step approach showed that the assumption of a constant hazard ratio was not valid for the used dataset. The results of the 1-step network meta-analysis could be presented as pooled parametric survival curves. CONCLUSIONS: Reconstructing data from published Kaplan-Meier curves allows for the inclusion of all relevant studies in a network meta-analysis, even if hazard ratios are not reported. Furthermore, it allows for network meta-analysis models that do not rely on the assumption of a constant hazard ratio, which have great value for cost-effectiveness models.