We investigated the importance of active, passive and snapshot exploration on spatial memory in a virtual city. The exploration consisted in traveling along a series of streets. 'Active exploration' was performed by giving directions to the subject who controlled his displacement with a joystick. During 'passive' exploration, the travel was imposed by the computer. Finally, during 'snapshot exploration', simple views of the scene were presented sequentially every 4 m. Travel velocity was the same in all cases. The three visual exploration modes were compared with three spatial memory measures: (1) scene recognition, (2) at the end of the path, reorientation toward the departure point and (3) drawings of the path shape. Scene recognition and estimation of the direction of the starting point of the path were not affected by the mode of exploration. In contrast, reproduction of the shape of the path was affected: the errors of reproduction were greater for the snapshot exploration than for the other two conditions; there was no difference between the other two conditions. These results suggest that (1) 2D image features from a visual scene are memorized. Moreover, (2) pointing towards the origin of the path relies on motion duration integration or a frame of reference integrated during displacement. Finally, (3) drawing the path shape involves a deliberate reconstruction process.