Nowadays, a huge volume datasets has been generated from numerical simulations and measurement equipments. However the classification and reference has been usually performed manually, such as vision or experience of a measurement engineer. There still has not been revealed efficient techniques for searching similarity between volumetric datasets.<BR>In this report, we propose a technique of performing the classification and reference using the critical point graph which expresses the feature of volume datasets in the simple figure. For this purpose, critical point graphs need to be generated efficiently and it is necessary to calculate the critical point in volume datasets efficiently.