Relay power allocation has been shown to provide substantial performance gain in wireless relay networks when perfect global channel state information (CSI) is available. In this paper, we consider a more realistic scenario, where such global CSI is subject to uncertainty, and we aim to design robust power allocation protocols for both the coherent and noncoherent amplify-and-forward relay networks. The problem formulation is such that the output signal-to-noise ratio is maximized under both the aggregate and individual relay power constraints. Our previous results show that these optimization problems can be formulated as quasiconvex optimization problems, and are solved using the bisection method via a sequence of conic feasibility problems. We extend these results to the case of uncertain global CSI, and design robust relay power allocations using the robust optimization methodology. For simple ellipsoidal uncertainty sets, the robust counterparts of these optimization problems are semi-definite programs and can be solved efficiently via interior-point methods.