We have developed an innovative procedure for model-based seismic inversion in areas with sparse or clustered and biased well control, where lithofacies and geobodies cannot be adequately sampled in wells and correctly represented in a priori acoustic impedance (AI) models constructed with conventional kriging methods. We have applied seismic sedimentology for facies mapping before the model building process. Lithology-calibrated seismic stratal slices contain rich information for analyzing sedimentary geomorphology and dispersal patterns of depositional systems, providing independent geologic knowledge to constrain a priori models. The new information can be incorporated by directional kriging that properly addresses facies types, orientations, and facies boundary conditions. Finally, the improved a priori model can be applied in Bayesian inversion for an updated inverted AI volume. This procedure was applied in a 3D project in Saidong Depression, Erlian Basin, China, with promising results, achieving inverted AI maps with a more complete facies representation, a more reasonable sediment dispersal pattern (orientation), and clearer facies boundaries.