Actually, we look for patterns of shared mutation using methods that have been tested on knowns:
Here is a hint - similarities are certainly informative, but it is the patterns of shared, unique characters that are indicative of descent. And this has, in fact, based on tested methods:
Science 25 October 1991:
Vol. 254. no. 5031, pp. 554 - 558
Gene trees and the origins of inbred strains of mice
WR Atchley and WM Fitch
Extensive data on genetic divergence among 24 inbred strains of mice provide an opportunity to examine the concordance of gene trees and species trees, especially whether structured subsamples of loci give congruent estimates of phylogenetic relationships. Phylogenetic analyses of 144 separate loci reproduce almost exactly the known genealogical relationships among these 24 strains. Partitioning these loci into structured subsets representing loci coding for proteins, the immune system and endogenous viruses give incongruent phylogenetic results. The gene tree based on protein loci provides an accurate picture of the genealogical relationships among strains; however, gene trees based upon immune and viral data show significant deviations from known genealogical affinities.
======================
Science, Vol 255, Issue 5044, 589-592
Experimental phylogenetics: generation of a known phylogeny
DM Hillis, JJ Bull, ME White, MR Badgett, and IJ Molineux
Department of Zoology, University of Texas, Austin 78712.
Although methods of phylogenetic estimation are used routinely in comparative biology, direct tests of these methods are hampered by the lack of known phylogenies. Here a system based on serial propagation of bacteriophage T7 in the presence of a mutagen was used to create the first completely known phylogeny. Restriction-site maps of the terminal lineages were used to infer the evolutionary history of the experimental lines for comparison to the known history and actual ancestors. The five methods used to reconstruct branching pattern all predicted the correct topology but varied in their predictions of branch lengths; one method also predicts ancestral restriction maps and was found to be greater than 98 percent accurate.
==================================
Science, Vol 264, Issue 5159, 671-677
Application and accuracy of molecular phylogenies
DM Hillis, JP Huelsenbeck, and CW Cunningham
Department of Zoology, University of Texas, Austin 78712.
Molecular investigations of evolutionary history are being used to study subjects as diverse as the epidemiology of acquired immune deficiency syndrome and the origin of life. These studies depend on accurate estimates of phylogeny. The performance of methods of phylogenetic analysis can be assessed by numerical simulation studies and by the experimental evolution of organisms in controlled laboratory situations. Both kinds of assessment indicate that existing methods are effective at estimating phylogenies over a wide range of evolutionary conditions, especially if information about substitution bias is used to provide differential weightings for character transformations.
In reality, the "junk code" contains a lot of very useful data in terms of evolution. Such DNA is under less constraint than is coding sequence, and so can accumulate more mutation, and this makes it easier for analytical programs to operate efficiently.
I know this because my line of research in graduate school specifically used noncoding DNA to look at phylogenetic relationships.