Understanding the accuracy of predicted peak pressure in wind engineering is crucial for effective risk assessment and structural design. This study investigates the capacity of large eddy simulation (LES) in predicting peak pressure coefficients on the flat roof and facade of a high-rise building, incorporating uncertainty quantification from 11-hour equivalent full-scale experimental measurements. The isolated and group configurations under 0° and 45° wind angles are analysed. The performance of two estimation methods for peak pressure coefficients – the traditional “epochal” approach and peaks over threshold (POT) – is evaluated, determining the optimal set of parameters that minimises uncertainty. While the stationarity test reaches satisfactory residuals for a 75-minute equivalent full-scale duration, 25- and 37.5-minute durations are also analysed. The traditional approach is preferred for its reliability and consistency, although POT notably excels in capturing higher peak pressure coefficients. Numerical results demonstrate negligible differences across all analysed durations and align well with experimental findings. A 25-minute equivalent full-scale duration is found to be a suitable representative of the entire signal. However, notable discrepancies near the upwind corners on the roof, with larger uncertainty in experimental data, underscore challenges in prediction accuracy for these regions. It emphasises importance of individual treatment and validation for each case.