PURPOSE: To provide a data sorting device to be functioned with high,'L accuracy at a high speed suitably for entire data and respective data distributions in the case of sorting over two kinds of data with different characters at least for each character.

CONSTITUTION: This data sorting device is provided with a learning data storage part 10, data selecting part 11 for reference vector to match the number of reference vectors and an initial value with learning data, reference vector storage part 12, closest reference vector search part 13, selecting part 14 for reference vector to be corrected, reference vector correcting part 15, closest information storage part 20, copy/erasure deciding part 21, copy/erasure part 22 and category deciding part 18. Further, this device is provided with a learning vector quantizing part 16 for performing learning with a learning vector quantizing algorithm, convergence deciding part 19 for accelerating learning by finishing the learning within the minimum number of times of learning for provided required accuracy, and estimating part 17.