This paper studies the use of amplify-and-forward (AF) relays for robust beamforming, with the aid of imperfect channel state information (CSI) at the sender. In particular, the maximization of the worst-case signal-to-noise ratio (SNR) at the destination terminal is studied under a bounded spherical region for the norm of CSI error vector. We show that under a condition on the estimated CSI quality, the robust-optimal collaborative-relay beamforming (CRBF) can be obtained by S-Procedure and rank relaxation techniques. In addition, a distributed algorithm is proposed by examining the structure of the optimal CRBF.