Abstract: In order to improve the accuracy of measurement of wide range water content in the process of crude oil dehydration. Method of a computation based on the electrical measurement of multiple modes is proposed. Through analysis of different models of oil water mixture, relationship between the pattern and dielectric constant as well as conductivity is established. The multiple modes of oil-water mixture are classified based on self-organizing neural network based on kernel method. The experimental results showed that the different modes of oil-water mixture can be identified by measuring the permittivity and conductivity. Therefore, the setting parameters in the water content analyzer will be revised in time, and the on-line measurement accuracy of the water content can be effectively improved.