Abstract A methodology devoted to the integration of seismic information within the geological stochastic modeling workflow has been developed in order to optimize the characterization of small scale internal reservoir heterogeneities. The workflow consists of three stages detailed here and in a companion paper in this volume ( Lerat et al. , 2006 ): Litho-seismic interpretation and definition of the 3D seismic constraint; geostatistical geological modeling using a 3D seismic constraint; and reconciling seismic data with the geological model. In this paper, we present the interpretation of seismic data in terms of 3D geological facies proportions at Girassol, a large and complex turbidite field located offshore Angola. This type of interpretation requires high quality seismic data, which was the case in the present case-study. Secondly, particular attention has been paid to calibration issues: for wells with available sonic data, vertical and lateral shifts have been determined at early stages of the analysis to ensure optimal local consistency between well logs and local seismic amplitudes. For the other wells, an optimal location has been found a posteriori by comparing electrofacies analysis results with geological facies proportions predicted from seismic. The proposed workflow integrates knowledge from various sources and with different spatial resolution and support: pre-stack seismic amplitudes and velocities, well logs, stratigraphy, and structural information. It is divided into three parts. Pre-stack stratigraphic inversion; Probabilistic seismic facies analysis from the inversion results based on discriminant analysis; and Computation of geological facies proportions from the previous results by a novel approach developed to account for scale differences between seismic facies and detailed geological facies description, in order to produce a geologically interpreted 3D model of the reservoir to be used in further steps of the reservoir modeling workflow.