In this study, multivariate data analysis, especially partial least squares regression (PLSR), is applied to analyze the near infrared absorbance spectra of fruit samples in order to acquire the inner qualities without destroying the samples. The calibration models have been established for the samples with raw data, first order derivative and second order derivative treatments, respectively. In the meantime, the models have been verified by using cross validation method. As anticipated, a model with higher correlation coefficient (r) and lower root mean square error of calibration (RMSEC) is preferred for both calibration and cross validation. The results reveal that the calibration models with second order derivative treatments have higher correlation coefficient, coefficient of determination, as well as lower RMSEC. Furthermore, the calibration models have been optimized by selecting partial wavelengths as new variables based on absorbance spectra and regression coefficient. The reasons why the calibration models are improved might be suitably cutting off partial wavelengths causing noises in the model.