In this paper we consider methods of gradient recovery in the context of primitive-variable finite-element solutions of viscous flow problems. Two methods are considered: a global method based on a Galerkin weighted residual procedure, and a direct method where gradients are recovered directly at individual nodes. The direct method has the benefit of utilizing the property of superconvergence as a natural consequence of its formulation, and furthermore requires no smoothing matrix to obtain the gradients at the nodal points. The two recovery schemes are considered with respect to two benchmark viscous flow problems of differing complexity. Both schemes are shown to produce comparable results, although the direct recovery method is found to be significantly more cost-effective than the global method.