The researchers demonstrated the application of backpropagation in neural networks for recognizing handwritten digits, specifically focusing on zip code digits provided by the U.S. Postal Service. The network architecture was specifically designed and constrained for this task, allowing the system to process normalized images of isolated digits with minimal preprocessing. This method achieved a notable performance, with a 1% error rate and about a 9% reject rate. The paper illustrated the effectiveness of using backpropagation for training neural networks in real-world applications, setting a foundation for future advancements in deep learning and image recognition technologies​.