Image classification is fundamental to many computer vision applications. The image feature set plays an important role in image classification. In this thesis, the objective is to investigate auto-correlograms for image classification problems. In a given image, the features are extracted and they are trained by the classifier. The feature calculation is based on the auto-correlograms approach. This work investigates the auto-correlograms approach for many classification scenarios. The auto-correlogram of the image captures the spatial correlation between similar colors in the corresponding image. Thus, the spatial nature of the auto-correlogram has a huge potential for image recognition and classification. Previously, this area was not thoroughly explored but it is believed that the nature of the auto-correlogram feature set can contribute to image classification. From Experimental results, we noted that the auto-correlogram has much better performance compared to the edge based feature set, for example the Gabor filters. However, we noted that the auto- correlogram has less performance compared to the color features, for example the color layout features. The experimental work in this thesis contributes to many fields in computer vision.