Emotional pictures are generally better remembered than neutral ones. However, in recent years, researchers have proposed that this emotional memory enhancement is not a unitary phenomenon but a compound process involving two opposite effects, especially when the memorized picture is a complex scene. On the one hand, emotional information may selectively enhance the memory of the emotionally arousing central item within the scene while impairing the memory of the peripheral background, leading to the “emotion-induced memory trade-offs”. On the other hand, emotion can enhance memory in a non-selective manner across central and background information, resulting in the “emotion-induced memory broadening”. Studies show that the occurrences of these two effects hinge on various factors related to stimulus properties and memory processes. As to the stimulus-relevant factors, memorized stimuli with negative and positive valences are more likely to drive the trade-offs and the broadening effects, respectively. Moreover, the intensity of the center-background association within the affective scene can also influence the selectivity of memory enhancement. Concerning the memory processes, task manipulations at the encoding (e.g., passive viewing vs. strategical processing), consolidation (e.g., sleep vs. no sleep), and retrieval (e.g., recognition vs. cued-recall) phases may differently affect the selectivity of emotion-induced memory enhancement. Hitherto, the cognitive and neural mechanisms underlying the emotional memory trade-offs and broadening effects remain unclear. There is a debate on whether the memory trade-offs effect is an automatic process independent of attention narrowing and post-stimulus elaboration. Besides, a few studies reveal that a core neural network involving the amygdala, hippocampus, fusiform, temporal pole, and inferior frontal gyrus is associated with the trade-offs effect, with the activation of some other brain regions dependent on valence and arousal levels of the stimuli. Further research needs to compare the mechanisms of the emotion-induced memory broadening with those of the trade-offs at both behavioral and neural levels, particularly taking a closer look at the automaticity of and the forms of memory representations involved in these effects. In addition, extending these effects from spatial to other (e.g., temporal) dimensions may help elucidate how emotional signals selectively enhance the memory of complex scenes from different perspectives.