Near infrared (NIR) spectroscopy technology based on a portable NIR analyzer, combined with PC-OSC algorithm, EOSC algorithm and generalized regression neural network (GRNN) has been applied to establishing quantitative models for prediction of acidity in 112 orange samples. The obtained results demonstrated that the fitting and the predictive accuracy of the models with EOSC algorithm were satisfactory and the EOSC algorithm was not as susceptible to overfitting the data as PC-OSC algorithm. The correlation between actual and predicted values of calibration samples (Rc) obtained by the EOSC acidity model was 0.888 0, and prediction samples (Rp) was 0.885 6. The RMSEP was 0.081 65. The results proved that the portable NIR analyzer combined with EOSC algorithm and GRNN can be a feasible tool for the determination of acidity in oranges.