Handwritten digit recognition helps make data entry quicker in places like banking and healthcare. The MNIST dataset includes 70,000 black and white images of numbers from 0 to 9, and it is frequently used to check how well different models can classify things. In this study, Convolutional Neural Networks (CNNs) were used, which are great at learning patterns from images on their own, so they need less help from people when getting the data ready. After resizing all the images to the same size and turning the digit numbers into a special format, they trained the proposed CNN models. These models achieved over 99% accuracy, which is better than older techniques like Support Vector Machines (SVM) and Multilayer Perceptrons (MLP). These results show that CNNs are dependable and work well in real-world situations. Newer approaches, such as combining quantum and classical networks or using ensemble models, are also being explored for improved performance.