A method and apparatus for classifying fingerprint image quality and a fingerprint image recognition system are provided. The method includes dividing a fingerprint image into a plurality of blocks, calculating and vectorizing quality classification parameters with respect to the respective blocks, obtaining quality classification values for the respective blocks by inputting the vectorized quality classification parameters into a trained quality classifier, and selecting a representative value of the quality classification values and determining the representative value as the quality of the fingerprint image. Since the quality of the fingerprint image can be accurately judged, reliability of the fingerprint image recognition system is increased.