With the increasing significance of rice as a staple food, particularly in Asia, the accurate classification of rice varieties has become essential. Traditionally, this task has depended on manual methods that are both error-prone and labor-intensive. This paper introduces an innovative solution utilizing machine learning and deep neural networks to tackle this challenge. Our study focuses on classifying five distinct rice varieties—Arborio, Basmati, Ipsala, Jasmine, and Karacadag—using image datasets. The dataset was carefully divided into training, validation, and test sets, comprising 70, 15 and 15% of the total images, respectively. Notably, each rice variety is evenly represented with 15,000 images per type, ensuring a balanced dataset for robust model development. Two primary deep-learning models were employed. The first model, a Vanilla Convolutional Neural Network (CNN), featured an advanced architecture with convolutional layers, batch normalization, and dropout layers. The second model utilized Transfer Learning, leveraging the VGG16 architecture with pre-trained weights. Both models exhibited exceptional performance on the test dataset, with the Vanilla CNN achieving an impressive accuracy of approximately 99.67%, and the VGG16-based model achieving an accuracy level of about 99.42%. Further optimization through fine-tuning of the VGG16-based model resulted in an accuracy enhancement to approximately 99.92%.