Organisations and end-users are increasingly using Cloud resources to take advantage of the anticipated benefits of a more cost effective and agile IT infrastructure. Virtual machines are provisioned based on a selection of available images, which often contain the operating system and the software stack required by applications. When managing an image library, some of the challenges faced by a resource provider include (i) identifying the optimal number of virtual machine images that satisfy most user requirements, and (ii) bundling software systems into images to minimise the time to provision virtual machines and ease the selection of images from an end-user's perspective. Using a traditional data centre workload, this paper proposes an algorithm for selecting software bundles for virtual machine images and examines the impact of bundle selection on the number and characteristics of resulting images. The main finding is that creating a small set of virtual machine images packed with the most popular software systems is enough to drastically reduce the time to deploy the software stack required by applications, and hence minimise the time for provisioning virtual servers in a Cloud infrastructure. © 2013 IFIP.