We consider a multiple input multiple output (MIMO) system wherein channel correlation as well as noisy channel estimation are both taken into account. For a given block fading duration, pilot-assisted channel estimation is carried out with the aid of a minimum mean square error estimator. The pilot sequences and data are jointly optimized to maximize the ergodic capacity while fulfilling a total transmit energy constraint. Using the optimization results, we study the effect of various system parameters and different types of channel knowledge. We show that systems with channel covariance knowledge and imperfect channel estimates can even outperform those with perfect channel estimation at the receiver, but without any channel knowledge at the transmitter. Further, we show that the signal to noise ratio and the block fading duration determine the qualitative behavior of ergodic capacity as a function of antenna spacing or angular spread. Finally, we find that the optimum pilot-to-data power ratio is always greater than one. It increases as the total transmit power decreases or as the block fading length increases.