The sensitive nature of legal data demands that legal AI focuses on privacy-preserving and decentralized learning approaches. Federated Learning (FL) has emerged as a promising method for enabling multiple participants to collaboratively train a shared model while safeguarding their sensitive information. Despite its potential, no prior work has explored the use of FL in legal NLP. To address this gap, we introduce , the new real-world FL benchmark for legal NLP, encompassing five legal NLP tasks and one privacy task derived from Chinese court data. Our comprehensive experiments highlight the unique challenges posed by real-world non-IID data in FL. This benchmark aims to drive further research on privacy protection in FL using real-world datasets, and model deployment in resource-limited environments. The code and datasets of FedLegal are available here .