Saccades, elicited by an identical visual stimulus in repeated trials, exhibit a certain amount of amplitude and direction scatter. The present paper illustrates how this scatter may be used to discern various properties of the subsystem that determines the metrics of a saccade. It is found in humans that scatter along the eccentricity axis is consistently more pronounced than along the direction axis. The ratio of amplitude scatter and direction scatter is approximately constant for all target positions tested. In addition, the absolute amount of scatter increases roughly linearly with target eccentricity but does not depend on target direction. We have explored whether these findings may reflect noisy variations in the neural representation of the saccade vector at the level of the collicular motor map. There are good reasons to assume that the motor map, at least in the monkey, (1) is organized in polar coordinates, (2) has a nonhomogeneous (roughly logarithmic) representation of saccade amplitude and (3) is anisotropic in nature (Robinson, 1972; Ottes, Van Gisbergen & Eggermont, 1986; Van Gisbergen, Van Opstal & Tax, 1987). To account for the intertrial variability in saccades, we have slightly extended an existing model for the collicular role in the coding of saccade metrics (Van Gisbergen et al., 1987) by allowing small variations in both the total amount and the location of the collicular population activity. We discuss how such noisy variations at the level of the motor map would be expressed in the metrics of saccadic responses and consider alternative models which could explain our data.