The prediction of pluvial flooding in urban catchment needs fast prediction methods due to short lead-times in rain prediction. As physically based models including pipe networks and surface flow are usually too slow for a real-time prediction, machine learning approaches could be used as a substitute. Such a machine learning approach based on artificial neural networks (ANN) was developed and tested in a prior study. The prediction model reads in time series of rainfall predictions and gives spatially distributed maximum water levels for this rain events within seconds. The database for the ANN training is generated with the physically based 1D/2D hydrodynamic pipe flow and surface model HYSTEM EXTRAN 2D (HE2D). To account for a variety of cases, a rainfall event catalog was built from the rainfall stations within 128 km radius from the study location, which served as an input for both HE2D and the ANN model. These events are of different durations, intensities and return periods, and their transferability to the study location is ensured by a regional frequency analysis. The aim of the present study is to investigate the uncertainty of pluvial flood prediction using an anonymous urban sub-catchment in the city of Hanover as test case. The underlying assumption of the study is that the most important source of uncertainty is due to the forecast of the storm events. An ensemble approach is used to account for uncertainty. The ANN is used for flood predictions, as the fast computation time makes an ensemble approach feasible. A K-Nearest Neighbour (KNN) model is used to select an ensemble of equivalent rain events from the database. Different metrics to quantify the rain events are compared. The ensemble of rain events generated with the KNN model is used as input for the ANN model. For testing the reliability of the ANN, the ensemble of dynamic water heights is compared to the reference results of the HE2D model. The spread of the maximum water level is investigated as a measure for prediction for the uncertainty of the forecast model.