The invention discloses a correlation clustered data slimming method based on a pre-generated tree structure. A cache table of a root correlation class node in the tree structure is initialized, and data meeting a preset slimming condition are input to the root node cache table; from the root node, slimming operation is carried out on each correlation class node in the tree structure according to breadth-first traversal; slimming result files corresponding to all target tables in the tree structure are obtained. According to the correlation clustered data slimming method, a data table is clustered according to service characteristics into the tree structure, the data characteristics are abstracted, correlation slimming is carried out on the data according to a correction key, and the data are slimmed under the condition that the original data service logic relationship is completely reserved.