The hypothesis that hippocampal activity at encoding is causally related to subsequent declarative memory expression is tested in the mouse, by using lidocaine inactivation of the hippocampus in combination with c-fos neuroimaging analysis. We employed a two-stage radial maze paradigm of spatial discrimination, which was previously shown to dissociate between declarative and nondeclarative expression of memory related to the same acquired material. In Stage 1 (encoding), mice learnt the constant location of food among a set of six arms (three baited, three unbaited) by being submitted repeatedly to discontiguous experiences with each arm separately ("go/no-go" discrimination). In Stage 2 (test-session), they are challenged with novel presentations of the arms, which are either combined into pairs of opposite valence ("two-choice" discrimination), or opened all six together ("six-choice" discrimination). Previous experiments have demonstrated that the "two-choice" situation is a critical test for declarative memory while "six-choice" discrimination may rely on procedural memory. We observed that (i) hippocampal activity measured by c-fos mRNA expression was increased by "go/no-go" learning, and this activation was blocked by pre-training local infusions of lidocaine; (ii) when performed just before each session of Stage 1, such inactivation spared the acquisition of "go/no-go" discrimination but produced, subsequently, a selective deficit in the "two-choice" test (not in the "six-choice" test). This study indicates that the hippocampus is "spontaneously" engaged in encoding processes necessary for long-term storage of discontiguous experiences under a form enabling flexible declarative memory expression.