This document proposes a method for optical character recognition (OCR) using neural networks and character classification based on symmetry. It involves three main phases: 1) A classification phase that categorizes training characters by their symmetry properties (vertical, horizontal, total, none), 2) A training phase that trains separate neural networks on each classified group, and 3) A recognition phase that preprocesses documents, extracts lines and characters, and recognizes characters using the trained neural networks. Experimental results show the method achieves up to 99.2% accuracy on Arial font without punctuation, outperforming traditional OCR methods. However, it requires 14.01 seconds for full network training.