Recent standardization of the MUMPS language (ANSI X11.1-1977) has resulted in increased investigations into its efficient implementation. Of particular concern is the method of implementation of the MUMPS hierarchical data file, or global. Two implementation techniques have found substantial support: a traditional technique utilizing linked physical blocks, and a more recent, self-indexing method based on the theory of balanced trees. Comparison of these methods shows that the balanced-tree implementation offers significant advantages in most MUMPS environments.