We investigated spontaneous variation of binocular torsion. Variation was expressed as SD of torsional eye positions measured over periods up to 32 sec. Subjects viewed a single dot target for periods of 32 sec. In half of the trials a large random-dot background pattern was superimposed on the dot. The movements of both eyes were measured with scleral induction coils. Spontaneous torsional movements were largely conjugate: cyclovergence was much more stable than cycloversion. This difference was not due to roll head movements. Stability of cyclovergence was improved by the background pattern. Although overall stability (SD of position) of cycloversion was unaffected by a background, the background induced or enhanced a small-amplitude torsional nystagmus in 3 out of 4 subjects. We hypothesize that the difference in stability of cycloversion vs cyclovergence reflects the greater importance of torsional retinal correspondence, compared to absolute torsional position. In two subjects we found evidence for the existence of cyclophoria, manifested by systematic shifts in cyclovergence caused by the appearance and disappearance of the background.