OBJECTIVES: Kaplan-Meier (KM) curves are commonly used to report time-to-event outcomes like overall survival (OS) and progression-free survival. For studies not explicitly reporting hazard ratio (HR) and confidence intervals (CI), KM curves can be utilised to estimate these summary statistics for conducting a meta-analysis. Here, we validate the method proposed by Parmar and colleagues for estimating HR (95%CI) by reading the KM curves. METHODS: Ten randomised controlled trials reporting HR (95%CI) and the associated KM curve for OS were randomly selected. Two independent reviewers read the survival probabilities from KM curves using an open source digitising software (Engauge digitizer). HRs for non-overlapping time intervals were calculated from the estimated survival probabilities and combined in a stratified way across time intervals to obtain an overall HR using the spreadsheet by Tierney and colleagues. The estimated HR was compared with the reported HR for each study. RESULTS: A mean error on the log scale of -0.001 (95%CI: -0.022, 0.019) was observed. This implies that by taking the exponentials, if the reported HR is 0.750, then the estimated HR would be 0.749. The 95%CI for the mean error spans zero indicating any systematic error is likely to be small and should not influence results in most analytic situations. Mean absolute error on the log scale was 0.027 (95%CI: 0.016, 0.037) indicating calculated HR lie within a factor of exponential (0.027) either side of the original value. No change in the direction of the treatment effect was observed in the estimated HR (95%CI) for any of the selected study. Reconstructed KM curves presented high accuracy and reproducibility. CONCLUSIONS: KM curves could be potential source of data and it is recommended that these should be used more frequently to estimate HR (95% CI), where not reported explicitly, for conducting meta-analysis in systematic reviews.