PURPOSE: To reduce the number of times for inputting / outputting data to a working file and to increase classifying speed when the data are classified by using a large capacity buffer as a key record stack.

CONSTITUTION: A record extraction part 111 reads an input record from an input device 10 and stores the input record to a record storing area 141. A classification key extraction part 112 prepares a key record and stores it in a key record storing area 142. When the storing area 142 is filled up, a key classification processing part 113 classifies this stored key buffer stack area 143 as one small train. When the stack area 143 is filled up, a key unification processing part 114 unifies this stored train as one big train and a record processing part 115 outputs the input records in the storing area 11 to the working file in the order of the key records. Thus, the number of times for inputting / outputting the data to the work file can be and throughput is improved.