Abstract. This paper presents a study on extreme precipitation using both stationary and non-stationary Generalized Extreme Value (GEV) models over a large number of samples distributed over Great Britain (GB) for the last century, aiming to gain insights in the spatial dependency of the GEV distribution. Not only L-Moments (LM) and Maximum Likelihood (ML) estimation methods but a Bayesian Markov-Chain Monte Carlo (B-MCMC) method are incorporated into the GEV models to characterize the uncertainty in the nonstationary risk-based assessment. The samples are generated using a toolbox of spatial random sampling for grid-based data analysis (SRS-GDA). The results show that a markedly large proportion (70 %) of the samples are favour nonstationary assumption GEV models as far as the annual maximum daily rainfall (AMDR) is concerned. The most frequent AMDR, as represented by the location parameter tend to be increasing over the time for more than half of the samples and in contrast, only 8 % have a downward trend. A spatially clustering pattern is also clearly present. For rarer (with 0.1 probability) AMDR, they are shown to have a tendency of becoming more extreme over time, for more than half of the samples. For the three methods, the LM method with stationary GEV maintain best results but for AMDR values with higher probability (5-year return level); the B-MCMC method with nonstationary GEV, however, outperform other combinations by a large margin for more extreme events (50-year return level). The findings suggest that an overhaul of the current engineering design storm practice may be needed in view of environmental change impact on natural processes.