The purpose of this particular work was to explore the benefits and drawbacks of sequential state updating for flood forecasting and identify factors or mechanisms affecting the updating process and thus controlling its performance. The Ensemble Kalman filter was employed to assimilate hourly streamflow observations into a simple but widely used conceptual rainfall-runoff model for flood prediction purposes. Ensembles were constructed by perturbing model forcing and parameters. Parametric perturbations were obtained from multiple model calibrations with an optimization algorithm. Errors in streamflow observations were characterized through an innovative yet simple empirical model. A sensitivity analysis was performed to evaluate the improvement of the first guess forecast. Additionally, the forecast skill was assessed as a function of lead-time. It was found that the improvement is mainly reflected in runoff volume, while the peak timecan be deteriorated as a trade-off of the assimilation process. Overall, ensemble-based models with sequential data assimilation outperformed the best-calibrated deterministic models for lead times of at least 1.5 days.