Hardness of dates is considered as one of the important quality parameters in the exportation. Visual inspection by manual graders is the current method commonly used for sorting dates in factories. Substituting this method by an automated technique will be more efficient. Therefore, the efficiency of NIR imaging technique was estimated to classify dates of Fardh and Khalas varieties based on their hardness. Histogram and textural features were extracted from NIR images and used in the classification models. Linear Discriminant Analysis (LDA) and Stepwise Discriminant Analysis (SDA) were used to determine the classification accuracies for the developed classification models (three classes model and two classes model) for both date varieties. Three classes model was developed using the three grades (hard, semi-hard and soft), while two classes model was developed using two grades (hard and soft) where soft and semi-hard were grouped together and treated as soft. For Fardh variety, 80.2% and 78.5% were correctly classified using LDA and SDA, respectively in three classes model. In two classes model, 95.7% were properly categorized using both LDA and SDA for the same date variety. However, only 61.8% and 59.8% were correctly categorized in the three classes model of Khalas variety by LDA and SDA, respectively. Similarly, 47.0% and 74.7% were properly classified in the two classes model using LDA and SDA, respectively for Khalas variety. The most contributing features to the discrimination were chosen to correlate with instrumentally measured features. Mean gray value, standard deviation and homogeneity were selected from Fardh variety images. Whereas, mean gray value, standard deviation and contrast features were selected from Khalas variety images. Poor correlation between instrumentally measured attributes and features extracted from NIR images was obtained in both varieties. Further investigation using other machine vision methods is required to improve the classification accuracies