Hyperspectral remote sensing is becoming a new tool for ecological research at large scales, and has been used to identify and quantify the mineral components of land surfaces and soils, and to estimate biochemical contents in plants, among other uses. In this study, the spectral characteristics of the salt marsh soils at Chongming Dongtan Nature Reserve in Shanghai were measured by a ground FieldSpecTM Pro JR spectroradiometer. The average spectral reflectance rates of all the band sets were extracted according to the band sets of Hyperion in EO-1 satellite. The partial least squares regression (PLS) was then used to predict the heavy metal contents in soils (with focus on Zn, Cr and Cu). The predicted values were compared with the measured values of these metals, obtained through laboratory analysis. The results showed that the correlation coefficients between the predicted and the measured values were 0.822, 0.761 and 0.775, and the average relative error for the Zn, Cr and Cu contents were 4%、3% and 4%, respectively. The results from this study indicated that the content of heavy metals in saltmarsh soils can potentially be quantitatively deduced from the reflectance rate. This would provide a basis for the use of hyper-spectral images for the evaluation and interpretation of heavy metal contamination in soils on a larger scale.