To find the sparse and symmetric n by n least-change secant update we have to solve a consistent linear system of n equations in n unknowns, where the coefficient matrix is symmetric and positive semidefinite. We give bounds on the eigenvalues of the coefficient matrix and show that the preconditioned conjugate gradient method is a very efficient method for solving the linear equation. By solving the linear system only approximately, we generate a family of sparse and symmetric updates with a residual in the secant equation. We address the question of how accurate a solution is needed not to impede the convergence of quasi-Newton methods using the approximate least-change update. We show that the quasi-Newton methods are locally and superlinearly convergent after one or more preconditioned conjugate gradient iterations.