Coherent multi-cell cooperative transmission, also referred to as coordinated multi-point transmission (CoMP), is a promising strategy to provide high spectral efficiency for universal frequency reuse cellular systems. To report the required channel information to the transmitter in frequency division duplexing systems, limited feedback techniques are often applied. Considering that the average channel gains from multiple base stations (BSs) to one mobile station are different and the number of cooperative BSs may be dynamic, it is neither flexible nor compatible to employ a large codebook to directly quantize the CoMP channel. In this paper, we employ per-cell codebooks for quantizing local and cross channels. We first propose a codeword selection criterion, aiming at maximizing an estimated data rate for each user. The proposed criterion can be applied for an arbitrary number of receive antennas at each user and also for an arbitrary number of data streams transmitted to each user. Considering that the resulting optimal per-cell codeword selection for CoMP channel is of high complexity, we propose a serial codeword selection method that has low complexity but yields comparable performance to that of the optimal codeword selection. We evaluate the proposed codeword selection criterion and method using measured CoMP channels from an urban environment as well as simulations. The results demonstrate significant performance gain as compared to an existing low-complexity method. Coherent multi-cell cooperative transmission, also referred to as coordinated multi-point transmission (CoMP), is a promising strategy to provide high spectral efficiency for universal frequency reuse cellular systems. To report the required channel information to the transmitter in frequency division duplexing systems , limited feedback techniques are often applied. Considering that the average channel gains from multiple base stations (BSs) to one mobile station are different and the number of cooperative BSs may be dynamic, it is neither flexible nor compatible to employ a large codebook to directly quantize the CoMP channel. In this paper, we employ per-cell codebooks for quantizing local and cross channels. We first propose a codeword selection criterion, aiming at maximizing an estimated data rate for each user. number of receive antennas at each user and also for an arbitrary number of data streams transmitted to each user onsidering that the resulting optimal per-cell codeword selection for CoMP channel is of high complexity, we propose a serial codeword selection method that has low complexity but yield comparable performance to that of the optimal codeword selection. We evaluate the proposed codeword selection criterion and method using measured CoMP channels from an urban environment as well as simulations. The results demonstrate significant performance gain as compared to an existing low-complexity method.