Understanding the principles behind how the cerebral cortex processes information requires identifying and characterizing its areal and laminar cytoarchitecture. In this context, deep learning offers a promising avenue to address these challenges by streamlining manual segmentation and offering observer-independent insights into cortical structure.We propose a deep-learning model for segmenting the cerebral cortex into areas and layers. We tested this solution on a dataset of one-dimensional cortical profiles derived from a non-human primate - common marmoset monkey (Callithrix jacchus) brain. The model was trained to recognize the layers in examples of agranular, dysgranular, and granular cortical areas with diverse, although clear, and agreed-upon laminar characteristics.Our findings demonstrate noticeably improved segmentation accuracy when the model is provided withan estimate of neuronal density and size in addition to the staining intensity profile,information on the cytoarchitectural type that the profile originates from, andweighted segmentation classes to balance the training dataset.Finally, we used the model explainability method (gradient class activation maps) to identify which elements of the input profiles cause the model to classify them into specific areas. We found that the criteria used by the model are consistent with the neuroanatomical guidelines found in the literature.Apart from being a tool for automated segmentation of the cortex into layers, the presented model can also provide valuable insight into the cytoarchitectonic properties of the primate cerebral cortex.Acknowledgments: NCN SONATA 2019/35/D/NZ4/03031.