Verification of ensemble streamflow predictions (ESP) requires large sample sizes to obtain accurate verification statistics. This means that verification of predictions for individual forecast locations and for specific forecast situations requires a retrospective verification approach. This involves simulating many ensemble streamflow predictions that would have been made in the past using the current forecast procedures. A retrospective approach to ESP verification is illustrated with a case study applied to the NWS BAYI4 forecast point in the Des Moines river basin. Different verification statistics to measure forecast resolution and reliability are examined for forecasts made at different times of the year, for different durations of the forecast window and for different forecast variables. Results show that predictability varies during the year and depends on the forecast variable at a given time.