Grid computing allows the use of distributed networks resources for tasks requiring a high processing power. A Grid infra-structure may help in the execution of these tasks and is able to coordinate their related activities, possibly regarding the provision of data files for the tasks executing in the grid nodes. GridFS is a system that enables data sharing in grid and heterogeneous distributed environments. By deploying servers over several nodes, it is possible to build a federated system integrating all local file systems and leveraging possibilities for tera-scale sized data storage. The proposed system was modeled and developed considering several aspects such as scalability, interoperability and performance. GridFS combines some characteristics from diverse file management systems, that is, GridFS provides an API for remote data access, copy operations allowing file transfers between servers, and some special features for grid environments. Apart from defining system characteristics and implementation aspects, this dissertation shows some experimental results about the service scalability and performance, and, as an evaluation, discusses the integration of GridFS with CSBase, a framework used to develop systems for grid computing.