Many studies have been done to investigate Naïve Bayes, the simple, yet effective technique for constructing classifiers. When modelling probability distribution of continuous variables with Naïve Bayes, researchers often assume that the predicting variables are normally distributed. However, this assumption is not intrinsic to the naïve Bayesian approach and may not hold for many situations. Therefore, a variety of nonparametric density estimation methods have been explored. In this paper, we are going to investigate the application of kernel density estimation to Naïve Bayes, and present experimental results of several artificial data sets.