In two experiments in a virtual pool the participants were trained to find a hidden platform placed in a specific position in relation to one (Experiment 1) or two (Experiment 2) objects; then, all the participants received a test trial, without the platform, and the time spent in the segment where the platform should have been was measured. In Experiment 1, groups differed in the distance between the landmark and the hidden platform. Test results showed that the control acquired by the landmark was different depending on its relative distance from the platform: Closer landmarks acquired a better control than distant ones. In Experiment 2, two objects, B and F, were simultaneously present during acquisition. Object B was just above the hidden platform (i.e., a beacon for the platform) while object F was above the edge of the pool (i.e., a frame of reference). On the test, the spatial location of B in relation to F was manipulated in the different groups and a generalization gradient was found: Participants spent more time in the segment where B was when B was in front of F (training position), and this time decreased symmetrically with distance of B from F. The two experiments provide convergent evidence of spatial learning effects in a virtual task with humans.