Extensive evidence indicates that the septum plays a predominant role in fear learning, yet the direction of this control is still a matter of debate. Increasing data suggest that the medial (MS) and lateral septum (LS) would be differentially required in fear conditioning depending on whether a discrete conditional stimulus (CS) predicts, or not, the occurrence of an aversive unconditional stimulus (US). Here, using a tone CS-US pairing (predictive discrete CS, context in background) or unpairing (context in foreground) conditioning procedure, we show, in mice, that pretraining inactivation of the LS totally disrupted tone fear conditioning, which, otherwise, was spared by inactivation of the MS. Inactivating the LS also reduced foreground contextual fear conditioning, while sparing the higher level of conditioned freezing to the foreground (CS-US unpairing) than to the background context (CS-US pairing). In contrast, inactivation of the MS totally abolished this training-dependent level of contextual freezing. Interestingly, inactivation of the MS enhanced background contextual conditioning under the pairing condition, whereas it reduced foreground contextual conditioning under the unpairing condition. Hence, the present findings reveal a functional dissociation between the LS and the MS in Pavlovian fear conditioning depending on the predictive value of the discrete CS. While the requirement of the LS is crucial for the appropriate processing of the tone CS-US association, the MS is crucial for an appropriate processing of contextual cues as foreground or background information.