We measured the amount of information transmitted by the direction of two-dimensional (2-D) arm movements of human subjects, and by neuronal populations in the motor cortex and area 5 of monkeys. We also compared the information transmitted by a motor cortical population when the predictability of the direction of movement was varied, i.e. when the target of the movement was the same, or different, in successive trials. The information transmitted by a neuronal population was measured using the "population vector" code (Georgopoulos et al. 1983, 1986) as the simulated directional output of the population. We found the following, (a) The information transmitted per movement increased in all cases with input information but more slowly than the maximum possible. No asymptote had been reached at 6.64 bits of input information; the best human performance at that level was 4.5 bits. (b) The average information transmitted by a motor cortical population (N=253 cells) was approximately 0.5 bits higher than that of best human performance at all levels of input information. These findings indicate that whereas information loss at the level of motor cortex, and during stages preceding it, increases with increasing input information, this loss remains constant at all levels of input information during processes intervening between motor cortex and movement. (c) The information transmitted by the neuronal population at a constant level of input information increased with the number of cells in the population and had not plateaued at the population size of 253 cells used in this study. (d) Consistently higher amounts of information were transmitted by motor cortical than by area 5 populations at all levels of input information (up to 7 bits) examined. The difference was approximately 0.16 bits. (e) Finally, the information transmitted by a motor cortical population was slightly higher (by 0.07 bits) when the movement was more than less predictable. Although this difference was small, it was consistent at all levels of input information (3-7 bits) and was statistically highly significant (p<0.005, paired t test). These results establish the information theoretical approach as a useful method for comparing (a) behavioral and neural data, (b) neural data between different brain areas, and (c) neural data obtained under different behavioral manipulations.