ABSTRACT This paper presents algorithms for generating idealized large-scale reservoir architecture representations. Discontinuous sands suspended in mud are discussed first. The accessible and floodable fractions of total pay are evaluated in 2-D and 3-D for specified conditions (well spacing, [N/G]-ratio,..). Next, a hybrid approach to the modelling of discontinuous permeability barriers such as shales is presented. Finally, stochastic modelling of faults, aquifers, and fractures is discussed. The authors conclude that a need exists for devicing approximate, yet realistic, computer-based methods for generating idealized reservoir representations for subsequent evaluation of the chance of success of proposed development schemes. Unless the reservoir architecture and hence the reservoir plumbing system is described in a quantitative (geometrical) manner, forecasting tools cannot be expected to yield realistic recovery factor estimates. The stochastic approach must always be guided by a geological/sedimentological model, as well as core, log, test, production, outcrop and seismic data.