At the moment, there is a vast amount of archival data spread across the Portuguese archives, which keeps information from our ancestors’ times to the present day. Most of this information was already transcribed to digital format, and the public can access it through archives’ online repositories. Despite that, some of these documents are structured with many plain text fields without any annotations, making their content analyses difficult. In this thesis, we implemented several Named Entity Recognition solutions to perform a semantic interpretation of the archival finding aids by extracting named entities like Person, Place, Date, Profession, and Organization. These entities translate into crucial information about the context in which they are inserted. They can be used for several purposes with high confidence results, such as creating smart browsing tools by using entity linking and record linking techniques. In this way, the main challenge of this work was the creation of powerful NER models capable of producing high confidence results. In order to achieve high result scores, we annotated several corpora to train our Machine Learning algorithms in the archival domain. We also used different ML architectures such as MaxEnt, CNNs, LSTMs, and BERT models. During the model’s validation, we created different environments to test the effect of the context proximity in the training data. Finally, during the model’s training, we noticed a lack of available Portuguese annotated data, limiting the potential of several NLP tasks. In this way, we developed an intelligent corpus annotator that uses one of our NER models to assist and accelerate the annotation process.