Few-shot class-incremental learning (FSCIL) performs classification continuously with limited new samples. Some mainstream methods constrain the model in incremental sessions or allocate spaces for incoming classes in the base session. However, these methods inevitably suffer from an overlap in the latent space of new and old classes, obstructing the discriminative ability of the model. To address that, we propose a novel two-stage FSCIL method using dual bridges containing sample bridge and memory bridge. In the base session, we train a model with the sample bridge, which is built by marriage of real samples, to preview knowledge. In incremental sessions, we align limited samples and the classifier with the help of memory bridge, which stores fixed feature centers. The dual bridges effectively link previous and subsequent sessions, minimizing the overlap. Extensive experiments on three publicly available datasets validate our method’s state-of-the-art performance.