Apples are crisp, sweet, and inexpensive. They have great economic value in China and are a characteristic pillar industry for rural revitalization in apple-producing areas. Moldy core disease is one of the main internal defects of apples, greatly affecting their quality. Failure to remove it promptly will be detrimental to apple branding and seriously affect the development of the apple industry. In this study, near-infrared spectroscopy was used to obtain the spectra of apples in different poses and illumination modes. Pretreatment methods such as detrending, baseline correction, and second derivative (SD) were used to pre-process the original spectra, and then a support vector machine (SVM) was used to establish a moldy core disease discrimination model for apples. Also, the effect of the illumination modes on mold core disease detection in different poses was analyzed, and the apple mold core disease detection method, by fusing the spectral information of different illumination modes, was investigated. The results indicate that in the single-illumination mode, the 180°illumination mode has a higher correct discrimination rate for regular apples, while the 135°illumination mode has a better correct identification rate for moldy core disease apples; for the single-illumination mode, the optimal SVM model is obtained in the rightward stance of the fruit pedicel and the 135° illumination mode. The models sensitivity, specificity, and correctness in the calibration set are 1, 0.978 2, and 0.986 3, and the sensitivity, specificity, and correctness of the prediction set are 0.888 8, 0.956 5, and 0.937 5, respectively. For the information fusion of different illumination modes, most of the information fusion of the illumination mode can improve the performance of the apple mildew discrimination model to some extent; the optimal SVM model is obtained in the information fusion of 180° and 135°illumination modes in the rightward stance of the fruit pedicel, the sensitivity, specificity, and correctness of the model in calibration set are all 1. The prediction sets sensitivity, specificity, and correctness are 0.888 8, 1, and 0.968 7, respectively. Compared with the optimal model of single illumination mode, the models performance is improved. This paper provides a new idea of rapid, nondestructive identification of apple moldy core disease by near-infrared spectra with the information fusion of different illumination modes, and the method can also provide a reference for nondestructive detection of the internal quality of other fruits.