This document summarizes research on recognizing handwritten ZIP codes in a postal sorting system. It discusses previous work on handwritten digit recognition using techniques like convolutional neural networks (CNNs) and analyzes the limitations of existing models. The document then describes a proposed system to identify handwritten ZIP codes on postal cards using a CNN trained on the MNIST database. It evaluates the system in terms of accuracy, sensitivity and specificity. Finally, it reviews several related studies and compares their network architectures, datasets and recognition accuracies. The overall goal is to develop an effective method for automatically sorting mail based on handwritten ZIP code identification.