Parkinson’s disease is a chronic neuro-degenerative condition that causes tremors, rigidity and bradykinesia, among other symptoms. These symptoms can manifest as alterations in handwriting, which are characteristic of Parkinson’s disease, resulting in distortions in the handwriting strokes while writing and also in disruptions in subjects’ handwriting fluency (i.e., degradation in the motor skills when writing). This work describes a study that aims to carry out the detection of Parkinson-produced alterations on parts of the handwritten texts, as well as which textual elements (e.g., letters, syllables, ...) are mainly Parkinsonian. For achieving such purpose, we have used a Convolutional Neural Network (CNN) and the Parkinson’s disease HandWriting (PaHaW) database. A collection of prediction experiments were carried out on different handwritten texts of this dataset, which have produced an average prediction accuracy results of over 65%.