Sorting is an important step in processing and packing lines of pomegranate fruits. Currently pomegranates are sorted into quality categories manually. But manual sorting poses problems such as tediousness, low accuracy, subjectivity etc. Moreover, manual sorting is not recommended for export quality fruits. Hence a machine vision system is required in order to sort the pomegranate fruits. The present paper is aimed at developing a robust non-destructive method to sort pomegranates using wavelet features and Artificial Neural Network (ANN) training. Pomegranates are sorted into ‘diseased’ or ‘healthy’ class. Initially, images of the diseased and healthy pomegranates are acquired from a local fruit market. As part of preprocessing, histogram equalization is applied followed by wavelet denoising. The preprocessed images are then fed to a feature extraction module where 15 spatial domain features and 252 wavelet features are extracted. Experiments were conducted to train ANN and calculate the performance based on spatial and wavelet features separately. Network performance is analyzed based on the parameters: Sensitivity, specificity, accuracy, mean square error and Receiver Operating Characteristic (ROC) curve. The results of experimentation showed that performance of ANN was high when wavelet features were used for training as compared to the spatial domain features.