The present invention relates to a method and network element for classification of bit strings such as data packets on a per-flow basis, wherein a plurality of classification fields are searched for in respective search trees based on a matching procedure in which an index value is obtained in a leaf node of a search tree for each classification field. Index values obtained in said searching step are used to derive a policy to be applied to said data packet, wherein the number of index values is reduced by combining intermediate results of the searching step or the deriving step. Thereby, computational efficiency can be improved and memory requirements reduced.