In multiuser multiple-input-multiple-output (MU-MIMO) systems, limited feedback schemes using a vector quantization codebook (VQC) for partial channel state information (CSI) can enable a base station (BS) to enhance the sum throughput at low feedback rates. However, the VQC has a limited throughput gain, although it requires a great deal of memory and complexity as the feedback rate increases. Therefore, we propose a high-precision feedback technique by using sparse approximation and compression called ¿compressive feedback (CF).¿ By using the sparse approximation with a unitary codebook and the compression via a ¿good¿ compression matrix, CF achieves greater sum throughput than VQC. Our analysis of the maximum achievable sum rate verifies the superiority of CF over VQC.