Although Federated Learning (FL) introduces significant advantages, it poses certain security risks. While original data is not directly disclosed, model parameters are still shared and may inadvertently reveal sensitive information using analysis techniques. Additionally, malicious clients can participate in the training process and submit erroneous model updates to compromise the system, potentially reducing the global model’s accuracy or disrupting the training process. This study will provide a blockchain-based FL solution to address these two concerns. Specifically, blockchain’s smart contracts and encryption algorithms are employed to ensure data privacy, manage access rights, and enforce client compliance with contractual terms. In addition, we develop a Multi-Krum-based algorithm to select and encourage the clients to submit reliable and accurate models, which prevents poison attacks by malicious users. Experimental results show that our proposed blockchain-based FL solution can maintain the model’s accuracy, which is always stable between 92% and 98%, while the conventional FL has a very low accuracy of less than 20%.