Since the extreme summer of 2003 the importance of early drought warning has become increasingly recognized even in water-rich countries such as Switzerland. Spring 2011 illustrated drought conditions in Switzerland again, which are expected to become more frequent in the future. Two fundamental questions related to drought early warning are: (1) How long before a hydrological drought occurs can it be predicted? (2) How long are initial conditions important for streamflow simulations? To address these questions, we assessed the relative importance of the current hydrological state and weather during the prediction period. Ensemble streamflow prediction (ESP) and reverse ESP (ESPrev) experiments were performed with the conceptual catchment model, HBV, for 21 Swiss catchments. The relative importance of the initial hydrological state and weather during the prediction period was evaluated by comparing the simulations of both experiments to a common reference simulation. To further distinguish between effects of weather and catchment properties, a catchment relaxation time was calculated using temporally constant average meteorological input. The relative importance of the initial conditions varied with the start of the simulation. The maximum detectable influences of initial conditions ranged from 50 days to at least a year. Drier initial conditions of soil moisture and groundwater as well as more initial snow resulted in longer influences of initial conditions. The catchment relaxation varied seasonally for higher elevation catchments, but remained constant for lower catchments, which indicates the importance of snow for streamflow predictability. Longer persistence seemed to also stem from larger groundwater storages in mountainous catchments, which may motivate a reconsideration of the sensitivity of these catchments to low flows in a changing climate.