Abstract: Today's large and high-resolution displays coupled with powerful graphics hardware offer the potential for highly realistic 3D virtual environments, but also cause increased target acquisition difficulty for users interacting with these environments. We present an adaptation of semantic pointing to object picking in 3D environments. Essentially, semantic picking shrinks empty space and expands potential targets on the screen by dynamically adjusting the ratio between movement in visual space and motor space for relative input devices such as the mouse. Our implementation operates in the image-space using a hierarchical representation of the standard stencil buffer to allow for real-time calculation of the closest targets for all positions on the screen. An informal user study indicates that subjects perform more accurate pointing with semantic 3D pointing than without.