The following passage from one of S. J. Gould's essays in the book Dinosaur in a Haystack may help:
"If doubts of close propinquity between hermit crabs and king crabs persisted, they were recently dispelled, and convincing new proof provided, in an elegant study published in 1992 ("Evolution of king crabs from hermit crab ancestors," by C. W. Cunningham, N. W. Blackstone, and L. W. Buss).
This study, done in the laboratory of my friend and colleague Leo Buss of Yale University, takes advantage of the revolution in taxonomy now underway thanks to recent technological advances that allow us to sequence DNA cheaply and rapidly (see the next essay). Conventional taxonomy struggles with fewer morphological, physiological, and behavioral traits often frustratingly subject to convergence. Sequencing of DNA and RNA provides hundreds or thousands of newly available characters (the ordering of strings of nucleotides, often highly conserved in evolution). These molecular data are, of course, equally subject to convergence and other forms of confabulation, but what a bounty of novel evidence!
Buss and colleagues sequenced part of an important gene that codes for ribosomal RNA, and they found 108 "phylogenetically informative" positions-an enormous increase in the number of useful characters for classification. They developed a matrix of similarities for all pairs of comparisons among twelve species of hermit and king crabs and a thirteenth more distant relative, chosen to anchor the tree (and called an "outgroup" in our jargon; they used the brine shrimp Artemia salina). They then applied a variety of standard tree-building techniques to this matrix of relative similarities. They achieved the same basic result with either of the two most common methods for tree-building: distance analysis, which works only with measured degrees of overall similarity; and parsimony, which constructs trees with a minimal number of evolutionary steps. The invariant result from different procedures greatly increases our confidence in their findings."
A third method, maximum likelihood, has become more popular recently. Promoted by Dr. J. Felsenstein, this method is said to be better than "traditional parsimony algorithms, which can give misleading results if rates of evolution differ in different lineages." (J Mol Evol. 1981;17(6):368-76)
As one can see from Gould's essay, nucleotide similarities are treated the same way as morphological, physiological or behavioral similarities, such as the absence or presence of a gular fold or legs, or the number of dorsal scale rows or ventrals etc. Some nucleotide similarities are shared ancestral characters (symplesiomorphies), while others may be convergences or shared derived characters (synapomorphies), while some are autapomorphies (unique changes within a species or lineage). It is up to the systematist doing the study to carefully sort out these different kinds of he similarities. If a mistake is made when choosing an outgroup, shared derived characters may be mistaken for shared ancestral characters and vice versa, resulting in a false hypothesis of branching order. For example, in one recent analysis of the Old World ratsnakes, Lampropeltis is choosen as an outgroup even though it is a taxon derived from a derived member of Elaphe. Not surprisingly, the species of Old World ratsnakes most closely related to Lampropeltis (i.e. E. scalaris) is found to be among the most basal species of Elaphe. When a different outgroup is used, for example, Ptyas, Elaphe scalaris is shown to be a more derived member of this clade.
Since many nucleotide substitutions are selectively neutral (for example, the "silent" substitutions of the third nucleotide in a codon which does not result in a change in the amino acid coded), there is much less of a chance that there has been convergent evolution due to adaptation to a similar environment (e.g. loss of limbs in snakes and legless lizards). Hence molecular analysis, though potentially more reliable and informative than morphological analysis, can nevertheless mislead if one is not careful about the choice of outgroups and/or the choice of characters.
