We present a new method to create stochastic models which honor both time-lapse seismic and well production data. A two-stage inversion workflow is used. The stochastic modeling allows us to estimate the uncertainty in the reservoir model and predict the realizations and production data. We firstly design three experiments which show the advantage of using time-lapse seismic data in the reservoir modeling and prediction. From these experiments, the inversion result using time-lapse data gives a better match to the actual saturation distribution. Finally, we describe and then present possible ways to address the uncertainty of in the required functions (variogram model, the relationship between porosity and permeability) used in our inversion approach.