Determination of mineral rock composition is an important part of formation evaluation in unconventional shale reservoirs. Two types of models are used for mineralogy modeling—deterministic and stochastic. Both models apply mathematical representations of the logging tool responses; however, stochastic modeling has become more popular due to its consideration of random distributions in the predictor and target variables. Stochastic mineralogy modeling algorithms usually produce solutions by minimizing a function reflecting the differences between the measured and modeled responses. However, due to the non-uniqueness inherent in inversion methods, the solution may not provide petrophysically meaningful results. To avoid producing compromised results, a method that incorporates geological constraints is proposed, in order to represent the geological and/or physical relations between the unknown parameters (inversion variables), leading to a meaningful mineralogy model.The proposed algorithm incorporates probability functions aiming to generate mineralogical solutions representing geologically and petrophysically sound results. The weight assigned to the penalties in the cost function depends on the probability function assigned to the constraints. A model is presented using the proposed algorithm: a pyrite-anhydrite constraint based on the iron and sulfur ratio. A field example from an Argentina well in the Vaca Muerta unconventional formation was processed using the proposed algorithm. The results using the proposed methodology show an excellent agreement with the available core measurements. These results demonstrate that the proposed method is an effective tool for improving integrated formation evaluation in challenging logging environments especially in unconventional reservoirs with complex mineralogy.