The two-dimensional phase unwrapping problem is studied. Using the minimum L^p-norm approach, we apply three different nonlinear multigrid algorithms for reconstructing surfaces from their “wrapped” values—two classical approaches and a novel Multilevel Nonlinear Method (MNM). The methods prove to be efficient even for difficult problems with noisy and discontinuous original images. The new method, MNM, exhibits the fastest convergence of the three for all the problems, given an appropriate choice of a damping parameter. Proposed methods for choosing this parameter automatically are mentioned.