An improved Kullback-Leiber distance is presented as a separable criterion for optimizing feature selection problems in pattern classification.A nonlinear parameter

Distributive Entropy of Euclidian Distance(DEED)

is introduced and based on which

the ratio of between-class DEED to within-class DEED (JRd) is defined as a criterion for the feature selection.DEED is a nonlinear measure for learning feature space

which gives the congregation and information measure of learning samples space.According to the result of Gaussian data experiments

it is concluded that the larger JRd be

the better separability of learning samples would be.