In this paper, we examine the effect of channel estimation errors on the performance of multiple-input-multiple-output (MIMO) systems that employ code-division multiple-access (CDMA) transmission. Channel estimation based on training techniques has widely been considered throughout the literature. However, employing these training techniques in MIMO-CDMA systems degrades system performance due to multiuser interference. This degradation is clear as the diversity advantage of the MIMO system diminishes with the increased level of interference. As a remedy to this problem, we propose a channel-estimation and data-detection scheme based on the superimposed training technique for space-time spreading systems. The proposed scheme enhances the performance of the space-time system by eliminating the interference effect from both the channel and data estimates using two decorrelators: channel and data decorrelators. We investigate the performance of the proposed estimation technique considering an asynchronous CDMA uplink transmission over frequency-selective slow-fading channels. In particular, we analyze the bit-error-rate (BER) performance of the multiuser system with two-transmit-antenna and V-receive-antenna configuration over Rayleigh fading channels. Compared with other conventional estimation techniques, our results show that the proposed estimation technique is more robust to channel-estimation errors. Furthermore, both simulations and analytical results indicate that full system diversity is achieved.