Many computational tasks require the determination of the Jacobian matrix, at a given argument, for a large nonlinear system of equations. Calculation or approximation of a Newton step is a related task. The development of robust automatic differentiation (AD) software allows for "painless" and accurate calculation of these quantities; however, straight forward application of AD software on large-scale problems can require an inordinate amount of computation. Fortunately, large-scale systems of nonlinear equations typically exhibit either sparsity or structure in their Jacobian matrices. In this paper we proffer general approaches for exploiting sparsity and structure to yield efficient ways to determine Jacobian matrices (and Newton steps) via automatic differentiation.