A few weeks ago I posted a question re: estimating divergence times. A number of people requested that I forward the responses. The most popular response was to use the program BEAST, which allows a relaxed molecular clock. Thank you to all who replied. You may want to try the software package BEAST which I am one of the authors of. It does Bayesian MCMC estimation of divergence times without assuming a particular tree topology. The paper to look at is: DRUMMOND, A.J., HO, S.Y.W., PHILLIPS, M.J., RAMBAUT, A., 'Relaxed phylogenetics and dating with confidence.' PLoS Biol 4, e88, 2006 http://biology.plosjournals.org/perlserv/?request=get-document &doi.1371/journal.pbio.0040088 There has been a workshop on dates and rates in Adelaide just last week, after which I have decided to use BEAST to calculate divergence dates. I have similar situation to yours: mtDNA, some estimates of divergence rates (from 2 to 12% per mln yrs), no calibration points, and rejected molecular clock. BEAST implements the model of "relaxed clock" which allows each branch on the tree have its own specific mutation rate. Given mutation rates (or strong mutation rate priors) it spits out an ultrametric tree with estimates of the age for major nodes and 95% credibility intervals around them (thus incorporating uncertainties in the phylogeny). Also, BEAST estimates parameters of population growth (if the model of growth is chosen by you as the most appropriate model for your organisms) and population size, because age of the nodes depends on pop size (tau=Ne*mu). You have to be patient with BEAST, though: there are thousands of models to navigate through, you have to have strong apriori priors and think of the best model for your data. Have you tried the program BEAST? That's what I used when I was in a similar situation. It incorporates uncertainty in your phylogeny by sampling over many different trees, weighted according to their posterior probability. It can give you an estimate, with a 95% confidence interval, of the divergence time between two or more groups. You can specify different evolutionary models (e.g. relaxed clock, or allowing different lineages to evolve at different rates, if that fits your data best). The program can be found at http://evolve.zoo.ox.ac.uk/beast/ Hope that helps. Sincerely, my advice is : do not try to calculate divergence times without good fossil calibration. Even if they are fashionable, phylogenetic methods cannot do better than fossils for estimating divergence times. So if you do not have any fossils, your estimations will have no meaning at all. You must be very careful when estimating divergence times with a single locus. Even if the gene is clocklike, what you will really be measuring is the locus specific divergence time, not the true species divergence time. This will vary throughout the genome due to stochastic processes governed by population parameters like population sizes, rates of divergence etc... How deep are the relationships you are trying to reveal? Species level, genus level, family level, etc? For just COI, you may have a difficult time getting good estimates of divergence, no matter what optimality criterion you use. I have used a simple 2% /million years rate before to get an estimate of divergence times, but this was for a small group of very recently derived species, and I was more interested in relative differences rather than absolute. A sweepingly broad estimate may be the best you can hope for given your poor resolution and single locus. You might check out "Estimating divergence times from molecular data on phylogenetic and population genetic timescales" by Arbogast et al 2002 for ideas. FWIW, in my research on Hawaiian bees I found that COI was evolving at a rate of up to 13% per million years (9% in *amino acids*!). That's a minimum rate (though based on the fastest-evolving species) calibrated by populations within a species on islands of known age. That of course is assuming that the bug arrived as soon as the island emerged, which is unlikely. Not that your things are necessarily that weird, but just pointing out that the real rate might be quite different from 1.4%, and to be wary if you're attaching absolute dates to it. Cheers Claire Claire McClusky Postgraduate Research School of Life and Environmental Sciences Deakin University Warrnambool VIC Australia clairefi@deakin.edu.au Claire McClusky