An algorithm is described that takes an ordered set of genomic mapping clones and fragment measurements from a restriction digest of each clone, and stochastically produces a contig restriction map of the clones. This is done by defining a map representation in terms of those pairs of fragments that coincide in the map. The fragment pairing is performed stochastically according to similarity between fragments. The search for good maps is guided by an objective function that uses a model based on minimum message length principles. Maps very close to the correct arrangement for known maps are found using this technique, without the need for the human supervision required by earlier techniques. Results from the application of the algorithm to a group of clones from the CEA gene region of chromosome 19 are given.