Distributed beamforming using multiple relay nodes is an effective means to provide power efficient transmission with diversity gain. Various approaches have been proposed to decide relay weights for each relay so that they can cooperatively relay signals. The maximum signal-to-noise ratio (MSNR) has been widely adopted as a performance measure in deciding relay weights. In this paper, we adopt the minimum mean square error (MMSE) to decide relay weights as the MMSE criterion can easily allow distributed implementation with local channel state information (CSI). We also discuss the channel estimation at each relay for distributed beamforming and the impact of the channel estimation error.