Because of the difficulty of performing full likelihood analysis over
multiple loci and the large numbers of possible orders, a number of me
thods have been proposed for quickly evaluating orders and, to a lesse
r extent, for generating good orders. A new method is proposed which u
ses a function which is moderately laborious to compute, the sum of lo
d scores between all pairs of loci. This function can be smoothly mini
mized by initially allowing the loci to be placed anywhere in space, a
nd only subsequently constraining them to lie along a one-dimensional
map. Application of this approach to sample data suggests that it has
promise and might usefully be combined with other methods when loci ne
ed to be ordered.