1. The binocular co-ordination of human horizontal saccades was analysed for the first time systematically over the full oculomotor range with a precise and accurate scleral sensor coil technique. Effects of amplitude (1.25-80 deg), direction (adduction vs. abduction and centrifugal vs. centripetal) and eccentricity (symmetrical about primary or between primary and secondary positions) were systematically investigated in three subjects). 2. To minimize extraneous effects of stimulus presentation on the programming of saccades, subjects were instructed to voluntarily change their gaze between two continuously visible targets. These were positioned on an iso-vergence locus, and thus contained no stimulus for disjunctive eye movements. 3. Under these conditions the amplitudes of the primary saccades of the two eyes were remarkably accurate; undershooting of the target by about 0.5 deg (independent of amplitude in the range 10-70 deg) was typical. This finding contrasts with the undershooting by about 10% described in the literature as characteristic for other stimulus conditions. 4. Saccadic peak velocities saturated at a mean asymptotic level of 502 +/- 32 (S.D.) deg/s for saccades of 40 deg and larger. The duration was linearly related to amplitude for saccades up to 50 deg; for saccades of larger sizes the duration increased progressively more steeply. Skewness values (acceleration time as a fraction of total saccadic duration) decreased from about 0.45 for saccades up to 10 deg to about 0.20 for saccades of 50 deg and larger. 5. Binocular saccades showed an abduction-adduction asymmetry and were not well yoked dynamically. The saccades of the abducting eye consistently had a larger size, a higher peak velocity, a shorter duration and were more skewed than the concomitant adducting saccades of the fellow eye. As a result, the eyes diverged transiently by as much as 3 deg during horizontal saccades. 6. Saccades also showed a marked centrifugal-centripetal asymmetry. Peak velocities of saccades towards the primary position were about 10% higher than peak velocities of corresponding centrifugal saccades. 7. These directional asymmetries were the main source of variability in the pool of saccades. In comparison, intra- and intersubject variability was minor in our sample. 8. Post-saccadic drift consisted of a vergence and a version component. The vergence component of this drift was a continuation of the vergence movement occurring during saccades. The version component, generally smaller than the vergence component, was directed towards the target position.(ABSTRACT TRUNCATED AT 400 WORDS)