Developing applications for distributed computation is becoming increasingly popular with the advent of grid computing. However, developing applications for the various grid middleware environments require attaining intimate knowledge of specific development approaches, languages and frameworks. This makes it challenging for scientists and domain specialists to take advantage of grid frameworks. In this paper, we propose a different approach for scientists to gain programmatic access to the grid of their choice. The principle idea is to provide an abstraction layer by means of a virtual file system through which the grid can be accessed using well-known and standardized system level operations available from virtually all programming languages and operating systems. By abstracting away low-level grid details, domain scientists can more easily gain access to high-performance computing resources without learning the specifics of the grid middleware being used. We have implemented such a virtual file system on HIMAN, a peer-to-peer grid middleware platform. Our initial experimental evaluation shows that the virtual file system only cause a negligible overhead during task execution.