Objects in real world are always located in certain context, which exert great influence on object perception. It is found that objects are named, categorized and searched more quickly and more accurately when foreground object and its background are semantically consistent than when they are semantically inconsistent. This phenomenon is called scene consistency effect. However, it is still controversial whether inconsistent scene is more attentionally attractive than consistent scene. Also, we know little about the effect of consistency on scene memory. In Experiment 1, participants were asked to freely view scene pictures. Meanwhile, eye movements were recorded. Result showed that inconsistent object was fixated earlier and for a longer fixation time, demonstrating that they are more attractive. In Experiment 2 participants demonstrated a similar eye movement pattern as in Experiment 1 when maintaining consistent or inconsistent scenes in short-term memory. In Experiment 3 a masked change detection paradigm was adopted to explore the effect of consistency on scene short-term memory. It was found that changes could be more accurately detected when scene consistency changes than when consistency remained during the task. This effect was robust since the same results were obtained when scene pictures were presented for a short period of time or in gray scale. Experiment 4 suggested that inconsistent scenes were more readily recognized in a long-term memory task. The present study explored the effect of consistency on both scene perception and memory, which contributes to revealing the mechanisms of scene processing.