A new operating system for the CRAY‐1 supercomputer implements four types of objects that can be manipulated by user tasks. A directory object is used to implement a hierarchical directory structure, a file object stores random access data, a pipe object permits intertask communication and a process object is used to execute user programs. The advantages of implementing a restricted number of object types in the kernel and the lessons learned from developing a supercomputer operating system from scratch are discussed.