We present a generative model to perform cosegmentation on an arbitrary number of images, where cosegmentation has been defined as the task of segmenting simultaneously the common parts between a pair of images. We build upon a previous work that introduced a new approach to model-based clustering under prior knowledge, and exploit its simplicity and flexibility to solve the problem of cosegmentation. We show experiments performed with datasets as diverse as slices of an MRI scan, frames from a video sequence, images in a database of objects, and with a set of 3D range images.