Animals and humans acquire knowledge through experience allowing them to flexibly generalize to novel situations. Several findings suggest that the Hippocampus (HC) is involved in the development of this conceptual learning not only for navigational but for other non-spatial variables if relevant to a task. However, how concepts arise with learning and how the hippocampus is involved in building these from the statistical structure of episodic memories still remains unknown. To study this, we trained mice on a binary categorization task consisting of a category split between low and high tone frequencies (Fig. 1A). We used 2-photon Ca2+ imaging to record CA1 excitatory population activity of HC in mice, at different phases in the learning curve, including sessions with new stimuli to probe generalization. (Fig. 1B,C). We found that mice do not generalize after first discrimination learning phase, which is surprising compared to what is expected from psychophysical experiments or simple machine learning classifiers (Fig. 1D). We performed a comparison of neural activity at the single and population level throughout the different phases of learning. Preliminary results show that selectivity to events is stable across the learning phases (Fig. 1E). Population level analysis with SVM decoders reveals encoding of task variables such as choice and outcome (Fig. 1F). Our results suggest that in mice, learning a linear category structure, does not arise from discrimination training but rather when single stimulus associations are not longer efficient, posing a challenge on storage and retrieval to support generalization.