This work studies the channel covariance matrix feedback problem in the coordinated multicell network, where a couple of base stations (BSs) collaborate with each other to transmit signals to the intended users. The design of transmit beamforming scheme requires the knowledge of channel state information (CSI), and the quality of CSI feedback will affect the system performance. We propose a novel channel covariance matrix feedback scheme based on matrix completion theory, in which the channel covariance matrix can be recovered with high accuracy via solving a nuclear-norm minimization problem, if a small number of covariance matrix entries contaminated by noise are fed back to BSs. Numerical results show that the proposed feedback scheme can yield high quality statistical channel knowledge at BSs.