The increasing global prevalence of diabetes mellitus has brought diabetic retinopathy (DR) to the forefront as a significant cause of visual impairment and blindness among adults. DR, a complication of diabetes, can be broadly categorized into Non-Proliferative Diabetic Retinopathy (NPDR) and Proliferative Diabetic Retinopathy (PDR), each comprising distinct clinical stages with varying severity. The accurate and timely detection and classification of these stages are essential for preserving vision, a task that grows in importance as the worldwide diabetes epidemic continues to rise, with an estimated 693 million cases projected by 2045. In response to this healthcare challenge, the development of efficient automated systems for DR diagnosis and classification has emerged as a critical research focus. Deep learning techniques, particularly Convolutional Neural Networks (CNNs), have demonstrated substantial promise in addressing this issue. This research presents a comparative evaluation of three prominent CNN architectures, namely Le-Net, AlexNet, and VGG16, in the context of DR classification, aiming to provide valuable insights into their performance. Among these architectures, VGG16 emerges as the standout performer, achieving an accuracy of 74.4%. This outcome underscores the potential of VGG16 as a robust tool for DR diagnosis, with its deeper and more intricate architecture proving adept at capturing intricate retinal features. As the research landscape evolves, future directions may encompass exploring newer model variations, incorporating more extensive and diverse datasets, fine-tuning hyperparameters, and enhancing the interpretability of deep learning models for real-world medical applications, ultimately propelling the field toward more efficient and accurate DR diagnosis and classification.