The hippocampus is a brain region involved in memory and spatial navigation. The dentate gyrus (DG), the first stage of hippocampal processing, sends information via mossy fibers to CA3 pyramidal neurons where it is integrated into a dense recurrent network. Yet, how these two hippocampal subfields encode information within the same task and how each restructures its coding with experience remain unclear. In our study, we trained mice in a virtual reality discrimination task based on olfactory and visual context cues. We recorded DG and CA3 activity in first-session and expert animals using in vivo electrophysiology and quantified the contribution of sensory, behavioral, and cognitive variables to neuronal activity with a Poisson Generalized Linear Model. We observed that in the DG, the capacity of single neurons to respond to multiple variables simultaneously, known as mixed-selectivity, increases with learning. Moreover, encoding of position, speed, and reward strengthens, revealing experience-dependent reorganization. In contrast, CA3 exhibits mixed-selectivity even before learning, indicating an intrinsic predisposition to integrate multiple signals. However, context, reward, and odors only become decodable in expert animals. These findings suggest that learning reorganizes DG and CA3 differently, enabling more specific encoding of key task elements. The DG builds its codes from experience, whereas CA3 refines and selects relevant signals on a preexisting framework.