Techniques for analyzing virtual machine images are described. In one embodiment, a subset of settings is extracted from one or more virtual machine images, the virtual machine images store therein values of the settings. The settings are used by software executing in virtual machines of the virtual machine images, respectively. A target one of the virtual machine images is selected and target values of the settings are obtained from the target virtual machine image. Sample values of the settings are obtained from a plurality of virtual machine images. The subset formed by identifying similarities and differences of the values between the virtual machine images.