We propose a 3D computational model for projective relations which is used in an integrated image and speech understanding system. The image and speech understanding system is being developed within a joint research project focusing on both technical and cognitive aspects of humancomputer interaction, Psychological experiments have been carried out to evaluate our computational model as an approximation of the meaning of projective prepositions used by humans in spoken instructions. These experiments investigate the acceptance of the model by subjects as well as the regularities regarding human usage of projective relations. Results of the computational model, the overall system, and the psychological experiments are presented.