In this paper, we compare several anisotropic adaptive strategies for RANS simulations : feature-based and goal-oriented approaches. If anisotropic mesh adaptation has proven its reliability for inviscid flows, additional challenges remain to be solved to have the full gain of adaptivity, including early asymptotic (spatial second) order convergence, early capturing of the scales of then physical phenomena, ... One of the key component in the adaptive process concerns the computation of the error estimate. We describe the standard multi-sacle L interpolation error estimate, also called features-based error estimate, and two goal-oriented error estimates : one based on viscosity solutions and a second approach designed for laminar flows. The study focuses on the (simple) Onera M6 wing where the low complexity of the geometry allows us to reach the asymptotic rate of convergence. To assess these results, a non-linear corrector is applied to the CFD solutions to provide point-wise error bounds on the given solution and quantities of interest.