An inverse scheme is developed for reservoir characterization using ensemble Kalman filter (EnKF) and non-parametric approach. EnKF has been studied by many researchers due to its novelties on recursive data processing, easy access to parallel processing, and quantifying uncertainties on its results. However, previous studies have shown poor characterization results with non-Gaussian permeability distributions. In this study, non-parametric approach is used to characterize permeability distribution with strong non-Gaussian characteristics. Normal score transformation is utilized to satisfy the Gaussian assumption of EnKF in the assimilation step. From the analyses of initial ensembles effects with non-Gaussian distributions, initial ensembles with higher similarity to the reference distribution would give more successful characterization results than those of less similarity. Additional improvement in reservoir characterization results is obtained by using normal score transformation.