Optical Character Recognition (OCR) serves as a pivotal technology, permitting the conversion of text within images or scanned documents into machine-readable and editable formats. While OCR has demonstrated success across various languages, tackling Arabic OCR remains a significant challenge due to the intricate nature of the Arabic graphical writing system. Among the most employed architectures in Arabic word recognition is one that combines Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Connectionist Temporal Classification (CTC). This architecture involves a decoding operation that transforms the network’s output into meaningful words. This paper utilizes IFN/ENIT database to assess and compare the results of three decoding methods, namely: Best Path Decoding (BPD), Token Passing (TP), and Word Beam Search (WBS). This aims to understand to what extent the decoding method can influence recognition results, then to identify the method that leads to high accuracy, specifically in the case of Arabic Handwriting.