In this paper we analyze different schemes for obtaining gradient estimates when the underlying function is noisy. Good gradient estimation is e.g. important for nonlinear programming solvers. As an error criterion we take the norm of the difference between the real and estimated gradients. This...