Abstract: The use of pictures in current user interfaces does not provide effective support for human computer dialogues,thereby restricting the communication capability in many cases.The layered model of picture semantics presented in the paper and the classification for the pointing actions adopted here provide a conceptual framework for approaching the problem.The system construction guidelines developed on this basis suggest that choices should be made for the layers and range in the picture semantics model according to the need of user’s task and be limited to layers as shallow as possible.A prototype human computer dialogue system constructed in the light of the guidelines is then described,which supports simple inquiry dialogues by using deictic and iconic pointing gestures and the visually presented semantics of a bitmap picture.