The aim of reflection synthesis of inserted objects is to generate reflections which would be seen by the same camera capturing the target scene and be reflected by the true reflective media in the target scene. This problem is inherently difficult because we are typically given only several, sometimes even just one, views of the target scene. In this paper, we explore the geometric constraints to synthesize geometrically correct reflections. We also demonstrate how to constraint the synthesized reflections to be photometrically consistent with those in the original target views. The proposed method, therefore, advances the current imagebased compositing techniques one step further toward the scenes with variations in both lighting conditions and viewpoints. We demonstrate our approach for real scenes.