This paper investigates the performance of different precoding schemes for a multiuser MIMO VLC system with channel estimation errors, an assumption that is commonly neglected in the literature. In particular, dirty paper coding, channel inversion, and block diagonalization, are considered for interference mitigation under imperfect channel state information. The impact of the variation of the beam angles of the transmitters and the field of view (FOV) of the receivers on the system performance is also examined. Simulation results reveal that, dirty paper coding provides the best performance under perfect channel state information (CSI). However, under imperfect CSI, suboptimal linear precoding schemes will give better performance. Furthermore, tuning the transmitting angles and the FOVs can significantly improve the system performance.