ABSTRACTSubsurface models of lithology are often poorly constrained due to the lack ofdense well control. Although limited in vertical resolution, high-quality threedimensional(3-D) seismic data usually provide valuable information regardingthe lateral variations of lithology. In this thesis, I will show how Bayesianapproach can be used to generate seismically constrained models oflithology. Unlike cokriging-based simulation methods, this method does notrely on a generalized linear regression model, which is inadequate whencombining discrete variables, such as lithology indicator; and continuousvariables, such as seismic attributes. This method uses a Bayesian updatingrule to construct a posterior probability distribution function of lithoclasses byusing a priori information from well data and the seismic likelihood to constrainthe 3-D geological scenarios produced by geostatistical technique, which isthen sampled sequentially at all points in space to generate a set ofrealizations. The realizations define alternative, equiprobable lithologicmodels. The methodology was applied to delineate productive reservoir zonein Boonsville, Texas. To achieve better result in the Bayesian SequentialIndicator Simulation, I used acoustic impedance obtained from a seismicinversion process as the attribute to constrain the simulation. It is expectedthat by using this attribute, the separation of the litho-class conditionaldistribution will be well defined and at the same time minimizing the overlapsbetween the two distributions. The lithology classification obtained from BSISis then integrated with the result of the seismic inversion to clearly delineatethe productive zone in the field.