Currently, inspection of wheat in the United States for grade and class is performed by human visual analysis. This is a time consuming operation typically taking several minutes for each sample. Digital imaging research has addressed this issue over the past two decades, with success in recognition of differing wheat classes and distinguishing wheat from non-wheat species. Detection of wheat kernel defects caused either by damage or disease has been a greater challenge. A study was undertaken using high-speed digital imaging to detect damaged U.S. grown kernels in freefall, one kernel at a time. The system is composed of hardware (camera, lighting, power supplies, and data acquisition card) and associated software for instrument control, data collection, and analysis. It was designed to capture images of freefalling kernels at opposing angles through the use of optical grade mirrors. Parameterization was performed on kernel morphological and textural characteristics of three views (primary and two reflections), whereupon these terms were used to develop classification models for sound and damaged classes. Fifty samples of hard red and white wheat subjected to weather-related damage during plant development were examined. Parametric (linear discriminant analysis, LDA) and non-parametric (k-nearest neighbor, KNN) classification models were tested to determine the image features that best fostered recognition of kernel damage (mold, pre-harvest sprouting, and black tip). The morphological features used in classification included area, projected volume, perimeter, ellipse eccentricity, and major and minor axis lengths. Textural features from calculated gray level co-occurrence matrices (including contrast, correlation, energy, homogeneity) as well as entropy were also considered, as were elliptic Fourier descriptors (truncated Fourier series functions that defined the contour of border in each view). The results indicate that with a combination of two morphological and four texture properties, classification levels attain 91–94% accuracy, depending on the type of classification model (LDA or KNN). The research findings are intended to lead to the streamlining of feature extraction in image-based grain inspection as well as to provide design criteria for high speed sorting.