This study aims to explore how supervised machine learning can be applied to the classification of handwritten signatures. The objective is to examine how accurately signatures can be classified and to evaluate the performance of different model architectures on this task, within the context of digitizing analogue information. A dataset of 2,293 handwritten signatures from 34 signers was collected from technical documents. The dataset was rebalanced and then preprocessed to reduce noise, normalize appearance, and standardize input dimensions. Eight models were designed, developed, tested, and evaluated using accuracy, precision, recall, and weighted F1- score. The results show that even simple models can achieve strong performance when preprocessed correctly and trained on a sufficiently large, balanced dataset. Performance varies significantly across architectures when evaluated on smaller datasets, where the number of training, validation, and test points is limited. Pretrained deep learning models, while more costly computationally, consistently outperform simpler approaches, making them a better choice for the task – especially when few data points are available.