As the volatility and randomness of the electricity price grows with the rapid development of China's power market, the need of electricity price correlation analysis emerges. To analyze the correlation between day-ahead power market features and electricity price, a feature perturbation correlation analysis method based on the multi-layer perceptron (MLP) is proposed. The MLP model is trained based on day-ahead power market operation data to capture the relativity pattern between features and electricity price. Perturbation is added to each feature of the original input dataset to generate perturbed dataset, and the prediction results of the perturbed dataset are obtained by the trained MLP. Correlation analysis conclusions are drawn through the comparison of perturbated features and corresponding electricity price output of MLP. The example analysis demonstrates that the proposed feature perturbation method is capable of analyzing non-linear correlation with multi-dimensional influence factors, and could provide deeper conceptions of the complicated correlation pattern between power market features and electricity price.