Abstract This paper proposes an indoor multiuser visible light communication (VLC) system using multiple lamps with orthogonal frequency division multiple access (OFDMA) techniques. For the users located in different places, the diversities of their channels help to optimally allocate the subcarriers to the users for high-speed transmissions. Considering the peak radiation power limited light sources, the modulation index on each subcarrier is jointly optimized with the subcarriers assignment scheme. This paper proposes a genetic algorithm and a two-stage searching method to solve this joint optimization problem. Considering the computational complexity, a distributed optimization scheme together with the resource block methods are applied. From simulation results, the distributed optimization method significantly reduces the computational complexity and only sacrifices about a 30% data rate. This paper takes the illumination requirements into account. By changing the additive DC bias, the average optical power is adjusted; then, the throughput performances over different illumination levels are discussed. Due to the channel feedback error, the channel uncertainty is modeled and considered in this paper. Based on the simulation results, knowing the channel uncertainty can provide around 6 times higher bit rate than not knowing the uncertainty when the normalized variance of channel uncertainty is −3 dB.