A new approach to describe the mesoscale variability caused by eddies in terms of the integral properties of all eddies contained in a certain ocean volume is suggested. For this purpose a pattern recognition algorithm to detect isolated anomalies in three-dimensional stratified data fields was developed and is introduced in detail. An application to 12 eddy-resolving density fields acquired in the framework of the German-Russian project MESODYN in four deep basins of the Baltic Sea, namely, the Arkona Basin, the Bornholm Basin, the Stolpe Furrow, and the Eastern Gotland Basin, demonstrates that the algorithm works properly. The most remarkable finding of this application is that the volume fraction occupied by the isolated anomalies or eddies is constant. It amounts to about 12% of the total volume for each data field no matter whether the stratification corresponds to a summer or a winter situation. Many possible applications of the pattern recognition are proposed to document its general usefulness. Finally, the main features of the algorithm and possible extensions are discussed extensively.