The invention discloses a self-organizing map neural network-based method for eliminating optical scanning holographic defocus noise, belongs to the fields of optical scanning holography and image noise elimination, and mainly solves the problem of defocus noise in the optical scanning holography. By virtue of the clustering analysis capability of the self-organizing map neural network, the defocus noise in reestablished images which are subjected to random encryption treatment can be eliminated. By virtue of the method, the problem of defocus noise in optical scanning holography can be solvedeffectively, and the noise eliminating method is applicable to each field.