Abstract: Abstract: Blueberry is an important fruit in the world, and its consumption has increased significantly in recent years because of its flavor and antioxidant capacity for anti-aging. In China, blueberry has increased greatly with rapid development of internet shopping and express. However, fruits with low firmness are easily to be damaged with juice leaking out, which is unacceptable by the consumer. Sugar content is also important to consumer. Accurate determination of blueberry quality is challenging since individual fruit are small and dark in color, and vary greatly in external and internal quality characteristics. Traditionally, blueberry quality was inspected by human in situ at the sorting line, which was inefficient and unreliable. Moreover, it's difficult to sort fruit by human based on sugar content and firmness, two quality attributes that are not only important to the consumer, but also directly impact the shelf life of blueberries. Therefore, hyperspectral imaging technique for predicting the firmness and sugar content of blueberries was researched. A pushbroom hyperspectral imaging system was used to acquire hyperspectral reflectance images from 490 'bluecrop' blueberries in two fruit orientations (i.e., stem and calyx ends) for the spectral region of 900-1 700 nm. Each fruit was then segregated by building a binary mask to recognize the fruit from the background using threshold segmentation in the hypercube. This was accomplished on the spectral image at 1 542 nm, which gave the maximum contrast between the fruit and the background. After that, each berry was identified by combining tilting and labeling operations on the masked image. From the regions of interest of each segmented blueberry image, mean reflectance was computed by averaging over all pixels. Finally, prediction models were developed based on partial least squares method using cross validation and were externally tested with 25% of the samples. Effect of fruit orientations on hyperspectral imaging prediction was evaluated by designated the spectral data in three treatment (stem end, calyx end, and whole fruit which averaged two mean spectra for the stem and calyx end regions to obtain on spectrum). Results showed that better firmness predictions (RC=0.911, RV=0.871) were obtained, compared to sugar content predictions (RC=0.891, RV=0.774). Fruit orientation had no or insignificant effect on the firmness and sugar content predictions. Further analysis showed that blueberries could be sorted into two classes of firmness. The result showed that hyperspectral imaging is promising for online sorting and grading of blueberries for firmness and sugar content as well.