The problem of causal joint multiuser detection and multichannel tracking (CJDT) in time-varying flat Rayleigh fading code-division multiple-access (CDMA) channels is considered. Starting upon the consideration that fading correlation prevents, in this scenario, the symbols to be decoded independently, the causal (i.e., corresponding to zero decoding delay) maximum-likelihood (ML) receiver is first derived: as expected, the resulting structure has a complexity which grows exponentially in the number of users, preventing any practical implementation. Sub-optimum detection strategies are thus investigated, basically through the expectation-maximization algorithm: the resulting receiver can be thought of as a combination of an interference suppressor and a Kalman filter, both operating iteratively. The results show that the newly proposed receiver achieves near-optimum performance and outperforms its direct competitors based on decorrelating detection