In this paper, we propose a feedback scheme of a channel covariance matrix with low overhead for massive multiple-input multiple-output systems. The proposed scheme decomposes the channel covariance matrix into the phase and amplitude parts. For the phase part, the element-wise uniform scalar quantization is performed. For the amplitude part, the following feedback information is generated: a bitmap which denotes a sign of difference between adjacent elements, the first value, an increment, and a decrement. To calculate the elements of the amplitude part, starting from the first value, when a bitmap is 1, the increment is added, otherwise the decrement is added. Simulation results show that the feedback overhead of the proposed scheme can be significantly reduced from 1.92% to 48.33% while the performance of mean square error can be maintained with that of the conventional scheme.