Abstract: It is a key issue for identifying crops under heavy-metal contamination on a large scale using satellite remote sensing data based on ground-sample spectral analysis model for evaluating crops with heavy-metal stress level. In this paper, hyperspectral data and leaf chlorophyll concentration of rice, heavy-metal concentration of soil were collected from three different polluted paddies in Changchun city, Jilin province, China, at mean time, Hyperion data were obtained. Spectral indices sensitive to heavy-metal contamination were selected by multiple stepwise regressions, and BP neural network models were created to estimate chlorophyll concentrations in rice under heavy-metal stress, which indicated the level of heavy-metal contamination. It was founded that an optimum ground-sample spectral analysis model was 4-11-7-1 network architecture with logsig thansfer function, and the classification accuracy for each pollution level was 100%. Moreover, it was successful to apply the ground-sample spectral analysis model to Hyperion data, and then achieve large-scale application in monitoring rice under heavy-metal contamination, the classification accuracy for each pollution level was more than 80%. This research may provide important references for large-scale application in the spectral model for assessing rice under heavy-metal contamination.