1. In early local feedback models for controlling horizontal saccade amplitude, a feedback signal of instantaneous eye position is continuously subtracted from a reference signal of desired eye position at a comparator. The output of the comparator is dynamic motor error, the remaining distance the eyes must rotate to reach the saccadic goal. When feedback reduces dynamic motor error to zero, the saccade stops on target. Two classes of local feedback model have been proposed for controlling oblique saccades (i.e., saccades with both horizontal and vertical components). In "independent comparator" models, separate horizontal and vertical comparators maintain independent representations of horizontal and vertical dynamic motor error. Thus, once an oblique desired displacement signal is established, the horizontal and vertical amplitudes of oblique saccades are under independent feedback control. In "vectorial comparator" models, output cells in the motor map of the superior colliculus act as site-specific vectorial comparators. For a given oblique desired displacement, a single comparator controls the amplitudes of both components. Because vectorial comparator models do not maintain separate representations of horizontal and vertical dynamic motor error, they cannot exert independent control over the component amplitudes of oblique saccades. 2. We tested differential predictions of these two types of models by electrically stimulating sites in the superior colliculus of rhesus monkey immediately after either vertical or horizontal visually guided saccades. We have shown previously that, despite the fixed site of collicular stimulation, the amplitude of the visually guided saccades systematically alters the amplitude of the corresponding component (horizontal or vertical) of stimulation-evoked saccades. However, in the present study, we examined the effect of the visually guided saccades on the amplitude of the orthogonal component of stimulation-evoked saccades. 3. For a fixed site of collicular stimulation, vectorial comparator models predict that the initial visually guided saccade will influence both components of the ensuing stimulation-evoked saccade via the single feedback comparator. By contrast, independent comparator models permit the independent manipulation of the horizontal and vertical amplitudes of these oblique stimulation-evoked saccades. 4. In total, we collected data from 15 collicular stimulation sites. Immediately after either horizontal or vertical visually guided saccades of different amplitudes, we measured the horizontal and vertical amplitudes of saccades evoked by stimulation of the intermediate or deep layers of the superior colliculus. For each site, the duration, frequency, and current of the stimulation train were held constant. 5. Under these conditions, stimulation-evoked saccades followed visually guided saccades with short latency (18.1 +/- 6.7 ms, mean +/- SD). For every stimulation site tested, although the amplitude of the component of stimulation-evoked saccades corresponding to the direction of the preceding saccade (horizontal or vertical) varied systematically, the amplitude of the orthogonal component was roughly constant. 6. Thus the horizontal and vertical amplitudes of oblique stimulation-evoked saccades can be manipulated independently. Moreover, the peak velocity-amplitude relationships, the instantaneous velocity profiles, and the ratio of horizontal and vertical velocities and durations were very similar to those of visually guided saccades. 7. Independent comparator models can readily account for the ability to manipulate the amplitude of one component of oblique saccades without affecting the other. However, two-dimensional local feedback models that cannot exert independent control over the horizontal and vertical amplitudes of oblique saccades should be carefully reevaluated.