Abstract: Parkinson’s disease is one the most common neurodegenerative diseases. This disease is diagnosed and controlled through clinical evaluations, which are considered imprecise and can lead to delayed and incorrect diagnoses. Therefore, a significant amount of research today focuses on finding methods to identify abnormal symptoms related to Parkinson’s disease through brain assessments and neuronal activity. Electroencephalography (EEG) is a non-invasive, cost-effective method with high temporal resolution for recording brain signals. Its applicability in the diagnosis of diseases and deciphering neuronal activity in humans has been proven many times. Many studies have analyzed these signals for the diagnosis of Parkinson’s disease using machine learning and deep learning methods. Despite significant advancements in artificial intelligence within the medical field, due to the inherently complex and non-interpretable nature of these methods, their usability has faced considerable limitations. Consequently, adding the capability of explanation and interpretation to deep learning models has become an important and challenging endeavor. One such interpretable model is the innovative SincNet neural network, which, in its first layer, uses the training of sinc functions to discover meaningful filters with clear physical meanings. In this research, we, for the first time, employ the interpretable SincNet network for the classification of Parkinson’s disease. After training the model using 10-fold cross validation approach, we validate the model and evaluate the learned filters in the SincConv layer. In the end, the model achieved an accuracy of 88.7%, demonstrating that the network inhibited delta and thera frequency bands and utilized alpha, beta, and gamma frequency bands for classification. These frequency bands had previously been reported as abnormal in Parkinson’s patients compared to normal individuals, highlighting that the high accuracy of the deep learning network’s diagnosis results from the extraction of relevant information rather than statistical bias.