High concentration of nitrate will cause many problems, such as water eutrophication and compromise of human beings’ health. A fast and stable approach was applied to predict nitrate concentration in solutions using the dual optical active correction continuous spectrum analyzer designed by our research group. Firstly, standard normal variate (SNV) was used to pretreat the spectral data. Then characteristic wavelengths of spectral curve were selected by using successive projections algorithm (SPA) and genetic algorithm (GA) respectively. Finally, partial least-squares regression (PLSR) was applied to build the spectral prediction model to predict nitrate concentration, and coefficient of determination (R<sup>2</sup>) and root mean square error of prediction (RMSEP) were introduced as the evaluation indicators of prediction models. For SNV-GA-PLS model, R<sup>2</sup>=0.9966 and RMSEP=0.1712, which outperformed than SNV-UVE-SPA-PLS model (R<sup>2</sup>=0.9896, RMSEP=0.3952). It demonstrated that he model which selects spectral characteristic wavelengths by GA can decrease the complexity of prediction model building and ensure the accuracy as well. Hence, SNV-GA-PLS model can be used to predict nitrate concentration in water with quick and steady performance.