The spectral data was collected by using two different infrared spectroscopies with 400 to 1 000 nm (visible-shortwave) and 1 000 to 2 500 nm (longwave) from yellow peach chips. Then four mathematic algorithms, i. e. standard normal variate transformation (SNV), multiplicative scatter correction (MSC), moving-average smoothing (MS) and 1st-derivative (1-Der), were utilized in data preprocessing. Regression models by linear partial least squares (PLS) and non-liner support vector machine (SVM) were constructed for the predicting the soluble solids content (SSC) and firmness in yellow peach chips, respectively. Moreover, the feasibility analysis for prediction of SSC and firmness were vitrificated by the external experiments. The results showed that the best performance for SSC prediction was obtained with R_p of 0.761, RMSEP of 1.998% and RPD of 1.532 by MSC-SVM algorithm in 400 to 1 000 nm. However, the best performance for firmness prediction was obtained with R_p of 0.862, RMSEP of 0.292 kg and RPD of 1.991 by MSC-SVM algorithm in 1 000 to 2 500 nm. All these findings demonstrated that the near-infrared spectroscopy could be utilized to monitor the quality of fruit chips with non-destructive attributes, and also positively promote the development of online automated grading system.