The clustering method based on resistivity values is a feasible method to identify the contaminated area from the monitoring results when using ERT to monitor site pollution. However, for layered soil media, when the resistivity value of a soil layer is close to that of contaminated soil, the method can not effectively distinguish between the soil layer and a contaminated area, resulting in inaccurate identification results. In order to solve this problem, a clustering algorithm based on covariance matrix is proposed in this paper. The horizontal and vertical covariance matrices are calculated respectively, and then, clustering is executed based on covariance matrices to identify the pollution area. The field experimental results show that the proposed algorithm based on covariance matrices can effectively overcome the clustering inaccuracy caused by soil medium stratification, the low resistivity soil layer caused by the infiltration of rain in the shallow layer was successfully avoided being identified as part of polluted area.