In this digital era, the demand for character recognition systems in various languages has been increasing steadily. Tamil is one of the oldest languages spoken till date and the need for efficient Tamil character recognition systems are now more than ever. This research work proposed a novel architecture of convolutional neural networks for the recognition of handwritten Tamil characters. The approach of using convolutional neural networks for the recognition of handwritten characters differs from other standard approaches in feature extraction. Proposed model was trained with a large dataset containing thousands of handwritten Tamil characters and has produced best results in both training and testing. Proposed trained model has also shown promising recognition rates when tested with different handwritten Tamil characters in both real-time and offline modes. The proposed work achieved a training accuracy of approximately 97%, which was significantly higher when compared to other approaches. © 2020 IEEE.