Restriction mapping is a process by which sites along an unsequenced DNA molecule are located through measurement of the distance between them. The authors consider a model for handling errors in restriction maps in which the error in measurement is modeled by using a normal distribution. Minimum message length techniques are applied to evaluate the relative merit of competing solutions to a restriction mapping problem. This is simplified, for cases where the cost of the model is equivalent, to a maximum likelihood criterion, which is used to rank multiple solutions by considering each arrangement as a different hypothesis. >