To investigate how visual and vestibular cues are integrated for the perception of gravity during passive self-motion, we measured the ability to maintain a handheld object vertical relative to gravity without visual feedback during sinusoidal roll-tilt stimulation. Visual input, either concordant or discordant with actual dynamic roll-tilt, was delivered by a head-mounted display showing the laboratory. The four visual conditions were darkness, visual-vestibular concordance, stationary visual scene, and a visual scene 180 degrees phase-shifted relative to actual tilt. Tilt-indication performance using a solid, cylindrical joystick was better in the presence of concordant visual input relative to the other visual conditions. In addition, we compared performance when indicating the vertical by the joystick or a full glass of water. Subjects indicated the direction of gravity significantly better when holding the full glass of water than the joystick. Matching the inertial characteristics, including fluid properties, of the handheld object to the glass of water did not improve performance. There was no effect of visual input on tilt performance when using the glass of water to indicate gravitational vertical. The gain of object tilt motion did not change with roll-tilt amplitude and frequency (+/-7.5 degrees at 0.25 Hz, +/-10 degrees at 0.16 Hz, and +/-20 degrees at 0.08 Hz), however, the phase of object tilt relative to subject tilt showed significant phase-leads at the highest frequency tested (0.25 Hz). Comparison of the object and visual effects observed suggest that the task-dependent behavior change may be due to an attentional shift and/or shift in strategy.