As a wildly used herbicide, Atrazine is mainly produced in China. In order to strengthen the routine detection of Atrazine exposure concentration and protect the health of occupational contact workers, it’s of great importance to develop on-site rapid detection method. A self-assembled near infrared spectrometer was used to record spectra of laboratory prepared atrazine solutions with concentration range from 10 to 1 000 mg·L-1. The influences of different pretreatment methods, such as multiplicative scatter correction, standard normal variate, first order derivative (D1), second order derivative and their combinations, different variable selection methods, such as competitive adaptive reweighted sampling (CARS) and genetic algorithm (GA), different regression methods, such as partial least square (PLS) and support vector regression(nu-SVR), on the model prediction accuracy were investigated. Results show that D1 is the best pretreatment method; GA obtain better results than CARS on selecting highly related spectral variables; nu-SVR model perform better than PLS model. The nu-SVR model constructed with 16 spectral variables selected by GA obtained the best results, whose coefficient of determination for calibration, the coefficient of determination for validation, root mean square error of calibration, root mean square error of validation (RMSEV) and residual validation deviation (defined as SD/RMSEV where SD denotes standard deviation) are 1, 0.99, 17.54 mg·L-1, 25.42 mg·L-1 and 11.43, respectively. These results indicate near infrared spectroscopy combined with chemometrics has great potential to quantify Atrazine concentration at workplace. This research explores the feasibility of quantification Atrazine at workplace with near infrared spectroscopy for the first time, which has great reference value for similar work in the future.