The feasibility of rapid recognition of an Hg-contaminated plant as a soil pollution indicator was investigated using near-infrared spectroscopy (NIRS) and chemometrics. The stem and leave of a native plant,Miscanthus floridulus(Labill.) Warb. (MFLW), were collected from Hg-contaminated areas (n1=125) as well as from regular areas (n2=116). The samples were dried and crushed and the powders were sieved through an 80-mesh sieve. Reference analysis of Hg levels was performed using inductively coupled plasma-atomic emission spectrometry (ICP-AES). The actual Hg contents of contaminated and normal samples were 16.2–30.5 and 0.0–0.1 mg/Kg, respectively. The NIRS measurements of impacted sample powders were collected in the mode of reflectance. The DUPLEX algorithm was utilized to split the NIRS data into representative training and test sets. Different spectral preprocessing methods were performed to remove the unwanted and noncomposition-correlated spectral variations. Classification models were developed using partial least squares discrimination analysis (PLSDA) based on the raw, smoothed, second-order derivative (D2), and standard normal variate (SNV) data, respectively. The prediction accuracy obtained by PLSDA with each data preprocessing option was 100%, indicating pattern recognition of Hg-contaminated MFLW samples using NIRS data was in perfect consistence with the ICP-AES results. NIRS combined with chemometrics will provide a tool to screen the Hg-contaminated MFLW, which can be potentially used as an indicator of soil pollution.