We present optimum bit and power loading algorithms for orthogonal frequency division multiplexing (OFDM) systems using an M-ary quadrature amplitude modulation (M-QAM) scheme. Several scenarios for the amount of channel state information (CSI) available at the transmitter are investigated. These include statistical, quasi-statistical, and instantaneous CSI. The optimality criterion used is minimizing the average/instantaneous total bit error rate while fulfilling total power budget and spectral efficiency constraints. Numerical examples show that optimizing the allocation of bits/power greatly enhances the system performance in the instantaneous CSI mode while not much gain is expected in either statistical or quasi-statistical modes. They also show that the system performance is more sensitive to bit allocation than to power allocation. Furthermore, it is shown that the quasi-statistical-CSI based loading yields the best performance given the complexity/performance tradeoff.