We have written DosFs, a new file system for UNIX that uses MS-DOS data structures for permanent storage. DosFs can be used anywhere a traditional UNIX file system can be used, and it can mount disks written by MS-DOS as regular UNIX partitions. DosFs can be used as the root partition, and exported by an NFS server. Our motivation for this work was efficient disk space utilization; DosFs provides from 10% to 13% better disk utilization than the 4.3 BSD Fast File System (FFS). In addition, we found that the disk block allocation algorithm used by DosFs lets us exploit large contiguous disk operations, providing a five-fold improvement over FFS for uncached operations. To the best of our knowledge, this is the first file system implementation that allows the user to specify the logical block size at mount time. A user can mount the same file system with a larger block size when file accesses tend to be sequential and a smaller one when they tend to be scattered. The MS-DOS structures were designed for a single-user system and do not support access control. We solved this problem with simple extensions that are backward-compatible with MS-DOS.