Twelve-month-olds realize that when an agent cannot see an object, her incomplete perceptions still guide her goal-directed actions. What would happen if the agent had incomplete perceptions because she could see only one part of the object, for example one side of a screen? In the present research, 16-month-olds were first shown an agent who always pointed to a red object, as opposed to a black or a yellow object, suggesting that she preferred red over the other colours. Next, two screens were introduced while the agent was absent. The screens were (1) red or green on both sides; (2) red on the front (infants’ side) but green on the back (the agent’s side) or vice versa; or (3) only coloured red or green on the front. During test, the agent, who could see only the back of the screens, pointed to one of the two screens. The results revealed that while infants expected the agent to continue acting on her colour preference and point to the red rather than the green screen during test, they did so in accord with the agent’s perception of the screens, rather than their own perceptions: they expected the agent to point to the red screen in (1), but to the green-front screen in (2), and they had no prediction of which screen the agent should point to in (3). The implications of the present findings for early psychological reasoning research are discussed.