Abstract: CARS was combined with SPA to select the important variables from the visible/near infrared spectrum of apple, then a variety of different modeling methods was used to develop calibration models for SSC of apple, finally, some comparative studies was done among those models. The analysis results showed that 31 variables which selected by CARS-SPA and PLS could build the most stable on-line detection model of apple soluble solids solids (SSC) , in this prediction model, the number of variables was only 1.69 percent of the orginal spectrum, the correlation coefficient of prediction and root mean square error of prediction were 0.936, 0.351% repectively. This study showed CARS-SPA could effectively extract important variables from spectrum of apple SSC, also it could simplify and improve the accuracy of prediction model effectively.