The k-NN classification algorithm has been broadly applied to text mining, pattern recognition and so on. Its
nearest neighbor k has directly effect on its classification accuracy. If the value of  k is too small, k-NN is sensitive to noise.
On the other hand, if  k is too large, its accuracy is also low. On this account, a fast  k value selection method is proposed.
Firstly, a candidate set of  k is calculated. Then an appropriate k is found quickly in this candidate set. Experiment results on
100 publicly available data sets show that, the proposed method can find an effective nearest neighbor successfully, which is
a method of good effect and potential.