A single algorithm is proposed for the numerical modeling of geological facies and corresponding petrophysical properties. All reservoir parameters, whether categorical such as lithofacies type or continuous such as permeability are coded and processed as a series of binary indicator variables. Stochastic modeling of the reservoir is based on Indicator Principal Component Kriging and the sequential simulation principle. The resulting alternative and equiprobable reservoir models honor the available information at wells and their statistics. These reservoir models are then used to investigate the impact of geological heterogeneities on flow performance prediction in a water-flood scheme. Analysis of various production parameters (cumulative oil production, cumulative water oil ratio and breakthrough times) indicates that they are more sensitive to the geological architecture of the reservoir than to details of the statistical distributions of petrophysical variables. Direct modeling of permeability across lithofacies, i.e. ignorance of the geological architecture, may lead to severe inaccuracy in reservoir performance prediction, particularly after the first transient years of production.