The visible/near infrared (NIR) technique has been applied in many fields because of its advantages of simple preparation, fast response, and non-destructive approach. Two hundred and ten fruits of six grades of pearl guava, a new popular variety in Taiwan, were randomly collected from a post processing center. Each fruit was scanned and measured at three detection positions to obtain its visible/NIR spectrum, sugar content, and hardness (bioyield force). Three calibration models named Multiple Linear Regression (MLR), Partial Least Square Regression (PLSR), and Modified Partial Least Square Regression (MPLSR) were tested, along with the first and second difference for spectra at 4 to 28 gap points. Results showed that the MPLSR model performed best in sugar content measurement, when it was applied with a three spectra per fruit strategy in the 400- to 2498-nm region. It had a coefficient of correlation r of 0.947 and standard error of predictions (SEP) of 0.721 degrees Brix. In hardness measurement, the PLSR model with a three spectra per fruit strategy in the 400- to 2498-nm region and pretreated with first difference of 28 gap points, reached the lowest SEP of 3.427 N. These results indicated the visible/NIR spectroscopy technique could be applied in sugar content and hardness measurement for pearl guava and reaches a validation accuracy of 94% in SEP when three spectra were included in the model and were pretreated by differential treatment with proper gap points.
