This article presents a new storage-retrieval method, called In-Deep Class Storage, designed for Flow-Rack AS/RS. If class-based storage is a well known method a lot studied in literature, this method is based on the statement that it is more efficient to dedicate the front layers of each bin to the class of the most popular items, rather than dedicating whole bins close to the drop-off station, as already studied in the literature. Obviously, this idea is not trivial to implement, due to the dynamic behavior of such racks. Thus, two separate algorithms were defined, one for storage, one for retrieval, enabling a dynamic use of our approach, with the only hypothesis of a Pareto distribution of the items demand. This article finally presents a simulation study, designed to compare the performance of random storage and retrieval to the use of the algorithms. This study shows significant improvement of the mean retrieval delay, main performance indicator chosen.