ABSTRACT: Process-Oriented Modeling (POM) is a technique that allows building high resolution 3D digital core models that honor sedimentary bedding. The core model is populated with petrophysical values to match porosity and/or permeability measurements from actual plugs. These models can be upscaled to a desired (e.g.: log-equivalent) support volume to supply reliable pseudo petrophysical values. Two possible applications of this technique to heterolithic reservoirs are here proposed: the first regards log-facies characterization, whereas the second regards permeability log estimation. Characterization of heterolithic log-facies. The use of conventional logs to define a facies zonation in geological formations made of alternating lithotypes whose thicknesses are below the vertical log resolution, carries out some lithologically mixed (heterolithic) log-facies. Petrophysical parameters from core plugs are unreliable to characterize these kinds of logfacies, due to their possible lack of lithological and statistical representativeness. 3D digital core models can be used to correctly upscale petrophysical parameters (porosity and permeability) from core scale to log-facies/cell-grid scale; the pseudo-parameters implicitly take in account the net-to-gross ratio. Permeability-log estimation. A moving-average-window technique can be applied to the whole 3D petrophysical core model in order to perform an upscaling to a log-equivalent support volume, so as to mimic the process of log acquisition along the well. Synthetic permeability curves can be estimated this way, and propagated to uncored wells through conventional logs processed by means of Artificial Neural Networks (ANNs), or any other supervised methods. The comparison between synthetic permeability curve estimated using POM-derived permeability as input to ANNs, and the synthetic permeability curve estimated using core plug measurements as input to ANNs, highlights that core-to-log scale-effects play a major role in the permeability estimation from conventional logs. The proposed technique is very effective in the preprocessing of core data, as it significantly reduces crossscaling problems related to the differences in core-log support volumes