Under the framework of the Kullback-Leibler (KL) distance, we show that a particular case of Gaussian probability function for feedforward neural networks (Nns) reduces into the first-order Tikhonov regularizer. The smooth parameter in kernel density estimation plays the role of the regularization parameter. Under some approximations, an estimation formula is derived for estimating regularization parameters based on training data sets. The similarity and difference of the obtained results are compared with other's work. Experimental results show that the estimation formula works well in the sparse and small training sample cases.