The partitionfinder group has a lot of discussion on all things related to this topic:

http://www.robertlanfear.com/partitionfinder/

http://groups.google.com/group/partitionfinder

There is no single correct answer for all studies.  If you are looking at the population structure of one species such as Homo sapiens, the answer is different than if you are studying how various tetrapod groups (amphibians, turtles, birds, snakes, etc) evolved from fish.

Hi Erika

I will assume you want to know the phylogeny of some organism and you have nuclear and mitochondrial data. In that case, as pointed out by Artur, you better split your dataset into nuclear and mitochondrial markers to obtain two phylogenies. And then you can compare what you obtain (similar branching, cytonuclear-disequilibria, etc.).

Usually it is not a good idea to combine the markers. One of the most important reasons is that inheritance might be uni versus bi-parental, which has important implications for the effective population size, Linkage is not important, as you also have independent markers among your nuclear genes.

in spite of this, there is some programs that might allow you to use both types of data in a single infile, given you tell the program which markers belong to each group. If I recall correctly, Migrate can be one of those.

The partitionfinder group has a lot of discussion on all things related to this topic:

http://www.robertlanfear.com/partitionfinder/

http://groups.google.com/group/partitionfinder

There is no single correct answer for all studies.  If you are looking at the population structure of one species such as Homo sapiens, the answer is different than if you are studying how various tetrapod groups (amphibians, turtles, birds, snakes, etc) evolved from fish.

I think there's another question in this question to think about: how to combine phased alleles? However for reconstruction of phylogenies on the level of different reproductively isolated species number of ambiguities will be small and will not influence results, in some cases, when there is high variation or hybridization it could be true nightmare. In addition Raxml treat ambiguities as additional state, so you can leave you sequences unphased.

Mitochondrial (mt) DNA has a very different evolutionary history due to the relatively rapid mutation/substitution rates (in animals), uni-paternal inheritance, and lack of recombination. As such it is often a bad idea to concatenate nDNA with mtDNA. My suggestion would be to make a phylogeny with each type of DNA separately and compare the resulting phlyogenies. If you have a small enough data set you could also run a Bayesian Concordance Analysis using programs like BUCKy to test for concordance between your loci, thus avoiding the issue of concatenation altogether (see link below). 

http://www.stat.wisc.edu/~ane/bucky/

Also as Olesandr noted for most phylogenetic software you do not have to phase your nDNA since most can recognize heterozygous sites.  

Hope this helps.