Focusing on 30 children aged 3 and 4, the effect of landmarks in reconstructing spatial array in accordance with viewer-centered representation was studied. Subjects were instructed to turn round and reconstruct a triangular array which is comprised of three objects. In the experimental condition, subjects constructed the triangular array in the space enclosed with opaque sheets to prevent the subjects from using a clue or a framework comprised of landmarks (such as windows, doors, and furniture), while those sheets were removed in the control condition. In result, when constructing, subjects confused with the objects on the left and the right (allocentric response) or maintained their positions (viewer-centered response), but the percentages of allocentric and viewer-centered responses between experimental and control conditions showed no effect of landmarks. Many of the subjects were very young, but, performed in accordance with viewer-centered representation independently of surroundings, and presented allocentric performance regardless of available landmarks presence. In conclusion, it was suggested that the allocentric response was not attributed to the direct association of an object position with a landmark, and that children aged 3 and 4 can show viewer-centered performance in spatial cognition. Finally, possible involvement of additional processes such as spatial updating and reference to special feature of anistropic space was discussed.