<p>The naïve Bayesian classifier (NBC) is based on an assumption that all attributes are independent with each other, and therefore, it is designed according to the estimation of marginal probability density function (PDF). However, this assumption leads to inability of NBC when conditional dependency information exists among attributes. To address this problem, this paper proposes a enhanced Bayesian classifier (EBC) for continuous attributes. To effectively use dependent information among attributes, a multivariate Gaussian kernel function is applied to estimate joint PDF of these attributes using multiple smoothing parameters. Based on this, there are two types of EBC: one uses a single smoothing parameter for the entire attributes, and the other utilizes reinforcement learning method to find smoothing parameter for each attribute. Extensive experiments are conducted to evaluate the performance of the proposed methods on several University of California, Irvine datasets and Tennessee–Eastman Process dataset with continuous attributes. The superior performance shows the effectiveness of EBC and indicates its wide potential applications in data mining.</p>