A method was designed to study frequency distributions of hydrologic variables (rainfall and discharge) which combined a stochastic model for hourly rainfall with a general conceptual model for transforming rainfall into discharge. The model generates many different flood events over a given simulation period to evaluate hydrologic risks. The Simulated Hydrographs for flood Probability Estimation (SHYPRE) method was based on the use of observations to describe phenomena and statistically reproduce them. Frequency distributions of hydrological variables are built empirically from simulated rainfall and flood events. Extrapolation of these frequency distributions toward rare frequencies is performed by generating very different events over a long simulation period rather than by directly fitting a theoretical probability distribution on observed values. This method yields an original estimation of flood quantiles from common to rare frequencies and provides complete temporal data about these floods. In addition, the approach supplies more stable estimates of flood quantiles than statistical distributions fitted on observed values, even for frequent events. This is due to a better integration of rainfall data and the parametric design stability of the two models (rainfall model and rainfall-discharge model).