Horizontal and vertical eye movements of ten human subjects were recorded with a scleral induction-coil technique during voluntary pursuit of sinusoidal, triangular and pseudo-random target motions of different frequency, amplitude and dimensionality upon a dark, diffuse or structured background. Data processing included separation of the composite eye movement into a cumulative smooth and saccadic displacement, computation of gain and phase of the composite and smooth eye movements with respect to the target movement and analysis of retinal position error. Pursuit eye movements were never completely smooth. Smooth pursuit gain was always lower than 0.95 and saccades were used to supplement the smooth eye movements in pursuing the target with the proper amplitude. The gain of composite eye movements was about unity for sinusoidal target motions and ramps; it exceeded unity for the highest frequency components in a pseudo-random motion. The gain of the smooth eye movements decreased monotonously whenever target velocity increased. It was higher for single sine waves than for a pseudo-random motion, however, with pseudo-random motion it was relatively higher for the higher frequency components. Phase lags were in general smaller for single sine waves than for pseudo-random motion, but for the latter a phase lead of the smooth component was consistently found for the lower frequency components. During pursuit of a rhomboid trajectory, the eye movements showed directional errors which are interpreted as anticipatory behaviour. The distribution of the retinal error was symmetrical around zero. Its standard deviation varied between about 0.2 and 1.3 degrees; it was about proportional to target velocity and inversely proportional to smooth pursuit gain. It was limited by the insertion of saccades which were in general corrective. The influence of a diffusely illuminated background was minimal. A structured background inhibited smooth pursuit in the horizontal direction by about 10% and in the vertical direction by about 20%. This deficit of smooth pursuit was fully compensated by the insertion of more saccades and had no consequences for the standard deviation of the retinal error. The type of structure of the background was only of marginal importance. Horizontal pursuit was in general slightly smoother and more precise than vertical pursuit.