The invention discloses a CSI feedback method for a large-scale MIMO system based on compressed sensing, and belongs to the technical field of compressed feedback of channel state information in a large-scale wireless communication system. The CSI feedback method comprises the steps that a mobile station MS calculates retention sparsity according to a preset energy factor of the system; the MS calculates a necessary number of measurements according to the retention sparsity, and calls a measurement matrix to compress downlink CSI needing to be fed back according to the necessary number of measurements; the MS converts the CSI needing to be fed back into bit stream and feeds back the bit stream to a base station BS, wherein the CSI needing to be fed back comprises a measurement signal, a partial supporting set and the sparsity; the BS demodulates the sparsity containing noise, the partial supporting set and the measurement signal according to the received bit stream, and calls an improved reconstruction algorithm to reconstruct the downlink CSI. By adopting the CSI feedback method in the scheme disclosed by the invention, the feedback overhead of the CSI feedback based on the compressed sensing can be effectively reduced, and the computational complexity of the CSI reconstruction algorithm on the BS side is reduced.