Saccades that consistently over- or undershoot their targets gradually become smaller or larger, respectively. The signal that elicits adaptation of saccade size is a difference between eye and target positions appearing repeatedly at the ends of saccades. Here we describe how visual error size affects the size of saccade adaptation. At the end of each saccade, we imposed a constant-sized error by moving the target to a specified point relative to eye position. We tested a variety of error sizes imposed after saccades to target movements of 6, 12, and 18 degrees. We found that the size of the gain change elicited in a particular experiment depended on both the size of the imposed postsaccade error and on the size of the preceding target movement. For example, imposed errors of 4-5 degrees reduce saccades tracking 6, 12, and 18 degrees target movements by an average of 18, 35, and 45%, respectively. The most effective errors were those that were 15-45% of the size of the initial target eccentricity. Negative errors, which reduce saccade size, were more effective in changing saccade gain than were positive errors, which increased saccade size. For example, for 12 degrees target movements, negative and positive errors of 2-6 degrees changed saccade gain an average of 35 and 8%, respectively. This description of the relationship between error size and adaptation size improves our ability to adapt saccades in the laboratory and characterizes the error sizes that will best drive neurons carrying the adaptation-related visual error signal.