A collection of a large amount of users’ data is required if training of machine learning models is to be done traditionally. Privacy concerns and security risks are thereby introduced in this process. Legal or ethical reasons restrict the sharing of data. Therefore, the failure of traditional machine learning methods is encountered in sensitive domains. These challenges are addressed by Federated Learning (FL). Machine learning in FL is done by collaborating with multiple clients across decentralized clients and thereby providing data privacy. Yet a critical challenge is to be faced remains: Privacy of data while updated weights are being aggregated and transmitted should be maintained. The movement of model updates from one server to another provides a high possibility of interception, thereby leading to an increased risk of data exposure. Encryption of data comes here to help; in particular, homomorphic encryption (HE) plays a vital role here. The model weights should remain encrypted throughout the processes (aggregation and transmission) for privacy concerns, and HE facilitates this. HE also blocks any unauthorized access to raw model updates (even from the central server itself). Despite these constraints imposed, any essential mathematical computations on the encrypted data can be performed without even decrypting it with the help of HE. This paper examines the collaboration of homomorphic encryption with federated learning. Our experimental results show that FL protected by HE achieves an accuracy that is comparable to the accuracy that can be achieved by traditional FL. This study highlights that employing HE for privacy-preserving FL is a crucial solution for sensitive domains like the healthcare sector.