A near infrared inspection system was built to automatically detect whole almond kernels with concealed damage at an inspection rate of 40 nuts/s. The inspection device detects transmitted light through whole almonds from six different near infrared LEDs using a sine wave modulation–demodulation scheme. Multiple linear regression and discriminant analysis to classify nuts as concealed damaged or undamaged was performed. A classification error rate of 14.3% on the validation set was obtained with discriminant analysis, and a classification error rate of 20.0% was obtained with regression analysis. Most of the incorrectly classified nuts were those where it was difficult to objectively determine if they were damaged or undamaged.