BGP is traditionally configured to implement traffic engineering objectives without considering potential network dynamics. This might result in undesirable traffic distribution when network failures occur. In this paper, we present algorithms for interdomain traffic engineering that achieve the interdomain traffic engineering objectives under network failures. That is, we aim to configure routing policies so that traffic is distributed evenly. More importantly, the configuration is robust in the sense that it is able to achieve the specified traffic engineering goals despite network failures. We first investigate the coarse-grained robust configurations. The derived configuration can achieve optimal robust traffic engineering objectives for most network failures. Further, we develop a greedy algorithm to derive robust BGP configuration for any traffic distribution and link capacities. We use simulations to evaluate the robustness of the derived BGP configurations by applying the algorithm to both transit and stub ASs under realistic traffic demands. Our results show that the derived BGP configurations can improve the default configuration significantly in terms of achieving the robust traffic engineering objectives. Furthermore, our algorithm achieves robust traffic engineering goals without diminishing other routing objectives.