The multistage Wiener filter (MWF) is an efficient technique for interference mitigation in CDMA systems. A limitation to its implementation, however, is that poor rank selection causes performance degradation. Methods exist to overcome this issue, but they are often computationally expensive, so a rank prediction technique is desirable. In this paper, we develop one such method based on the number of codes that correlate with the code of interest above a set threshold, obtained by experiment. The prediction technique is validated by simulation under heavy loading, in severe multipath, and with moderate Eb/No common in cellular systems. The special cases of orthogonal users and two non-orthogonal users are considered to provide further insight into rank selection and validate the proposed algorithm. It is shown that the technique is able to accurately estimate the optimum rank, defined as the lowest rank at which the best performance is achieved, even in the presence of channel estimation errors, which are typically small (< 1 dB) because of the high power pilot. The analysis is presented for synchronous CDMA systems, but can be extended to asynchronous systems.