THE RETRIEVING PROCESS OF A CASE (OR A SET OF CASES) FROM THE CBR SYSTEM MEMORY STRONGLY DEPENDS ON THE CASE LIBRARY ORGANISATION. HIERARCHICAL MEMORIES ARE VERY EFFECTIVE IN TIME RETRIEVAL BECAUSE ONLY A FEW CASES ARE CONSIDERED FOR SIMILARITY ASSESSMENT PURPOSES, AFTER A PRIOR DISCRIMINATING SEARCH IN THE HIERARCHICAL STRUCTURE. MISSING INFORMATION REPRESENTS A MAIN DRAWBACK WITHIN DISCRIMINATION TREES OR NETWORKS ORGANISATION. ANOTHER MAIN PROBLEM IS THAT SOMETIMES IS IMPOSSIBLE TO REACH OPTIMAL CASES DUE TO AN EXPLORATION IN A WRONG AREA OF THE HIERARCHY. THIS LAST PROBLEM IS ADDRESSED IN THIS PAPER. THIS PROBLEM COULD BE DUE TO THE FACT THAT THE HIERARCHY OF NODES DOES NOT CORRESPOND TO THE RELLEVANCE OF THE FEATURES, AS FOR EXAMPLE A BAD-UNIQUE- DISCRIMINATION ORDERING OF THE ATTRIBUTES. OUR PROPOSAL TO OVERCOME THIS PROBLEM IS TO SPLIT THE CASE LIBRARY IN A SET OF DIFFERENT SMALLER CASE LIBRARIES. THUS, THE RETRIEVAL RPOCESS STARTS MACHING THE CURRENT CASE AGAINST A SET OF PROTOTYPE CASES, CALLED META-CASES, TO SELECT ONE MORE CASE LIBRARIES TO SEARCH IN. THIS APPROACH TRIES TO BUILD SEVERAL HIERARCHICAL ORGANISATIONS FOR DIFFERENT KIND OF CASES, MAKING THE CBR SYSTEM MORE ACCURATE AND RELIABLE. SOME PRELIMINARY EXPERIMENTAL TESTS ON REAL DATA ARE PRESENTED AND DISCUSSED TO SHOW THE APPROACH.THE RETRIEVING PROCESS OF A CASE (OR A SET OF CASES) FROM THE CBR SYSTEM MEMORY STRONGLY DEPENDS ON THE CASE LIBRARY ORGANISATION. HIERARCHICAL MEMORIES ARE VERY EFFECTIVE IN TIME RETRIEVAL BECAUSE ONLY A FEW CASES ARE CONSIDERED FOR SIMILARITY ASSESSMENT PURPOSES, AFTER A PRIOR DISCRIMINATING SEARCH IN THE HIERARCHICAL STRUCTURE. MISSING INFORMATION REPRESENTS A MAIN DRAWBACK WITHIN DISCRIMINATION TREES OR NETWORKS ORGANISATION. ANOTHER MAIN PROBLEM IS THAT SOMETIMES IS IMPOSSIBLE TO REACH OPTIMAL CASES DUE TO AN EXPLORATION IN A WRONG AREA OF THE HIERARCHY. THIS LAST PROBLEM IS ADDRESSED IN THIS PAPER. THIS PROBLEM COULD BE DUE TO THE FACT THAT THE HIERARCHY OF NODES DOES NOT CORRESPOND TO THE RELLEVANCE OF THE FEATURES, AS FOR EXAMPLE A BAD-UNIQUE- DISCRIMINATION ORDERING OF THE ATTRIBUTES. OUR PROPOSAL TO OVERCOME THIS PROBLEM IS TO SPLIT THE CASE LIBRARY IN A SET OF DIFFERENT SMALLER CASE LIBRARIES. THUS, THE RETRIEVAL RPOCESS STARTS MACHING THE CURRENT CASE AGAINST A SET OF PROTOTYPE CASES, CALLED META-CASES, TO SELECT ONE MORE CASE LIBRARIES TO SEARCH IN. THIS APPROACH TRIES TO BUILD SEVERAL HIERARCHICAL ORGANISATIONS FOR DIFFERENT KIND OF CASES, MAKING THE CBR SYSTEM MORE ACCURATE AND RELIABLE. SOME PRELIMINARY EXPERIMENTAL TESTS ON REAL DATA ARE PRESENTED AND DISCUSSED TO SHOW THE APPROACH.