Uncertainty in the calibration of the generalised Pareto distribution (GPD) to rainfall extremes is assessed based on observed and large number of global climate model rainfall time series for nine locations in the Lake Victoria basin (LVB) in Eastern Africa. The class of the GPD suitable for capturing the tail behaviour of the distribution and extreme quantiles is investigated. The best parameter estimation method is selected following comparison of the method of moments, maximum likelihood, L-moments, and weighted linear regression in quantile plots (WLR) to quantify uncertainty in the extreme intensity quantiles by employing the Jackknife method and nonparametric percentile bootstrapping in a combined way. The normal tailed GPD was found suitable. Although the performance of each parameter estimation method was acceptable in a number of evaluation criteria, generally the WLR technique appears to be more robust than others. The difference between upper and lower limits of the 95% confidence intervals expressed as a percentage of the empirical 10-year rainfall intensity quantile ranges from 9.25 up to 59.66%. The assessed uncertainty will be useful in support of risk based planning, design and operation of water engineering and water management applications related to floods in the LVB.