The rate of convergence of the Balancing Domain Decomposition method applied to mixed finite element discretization of second order equations is analyzed. The Balancing Domain Decomposition method, introduced by Mandel in [24], is a substructuring method that involves at each iteration the solution of a local problem with Dirichlet data, a local problem with Neumann data, and a "coarse grid" problem to propogate globally and to insure the consistency of the Neumann problems. It is shown that the condition number grows at worst like the logarithm squared of the subdomain size to the element size, in both two and three dimensions and for elements of arbitrary order. The bounds are uniform with respect to coefficient jumps of arbitrary size between subdomains. The key component of our analysis is the demonstration of the equivalence of the norm induced by the bilinear form on the interface and the H^1/2-norm of the interpolant of the boundary data. Computational results from a message passing parallel implementation on a INTEL-Delta machine demonstrate the scalability properties of the method and show almost optimal linear observed speed-up for up to 64 processors.