Genome-scale metabolic models for new organisms include thousands of reactions that are generated automatically: by inferring them from databases of reactions and pathways, existing models for similar organisms, etc. This process includes several iterations of the draft model analysis, error detection, and improvement; starting from more general issues and going deeper into details. Especially in the first iterations model evaluation by a human expert is important. But genome-scale models are targeted for computer simulation and analysis, and are too detailed and complicated to be easily understood by a human. For example, in the beta-oxidation of fatty acids pathway, a reaction missing for a particular fatty-acyl-CoA type (e.g. decanoyl-CoA) is a more specific issue than missing a whole enoyl-CoA hydratase step (which could happen if the corresponding enzyme is not found). But the abundance of reactions in the model (e.g. those corresponding to other beta-oxidation steps and presented for each of the different types of fatty-acyls-CoA) may hide this fact from the human. That is why we developed a method for knowledge-based scaling of metabolic models, providing a higher-level view of a model, keeping its essential structure and omitting the details.