Despite the widespread perception that evolutionary inference from molecular sequences is a statistical problem, there has been very little attention paid to questions of experimental design. Previous consideration of this topic has led to little more than an empirical folklore regarding the choice of suitable genes for analysis, and to dispute over the best choice of taxa for inclusion in data sets. I introduce what I believe are new methods that permit the quantification of phylogenetic information in a sequence alignment. The methods use likelihood calculations based on Markov–process models of nucleotide substitution allied with phylogenetic trees, and allow a general approach to optimal experimental design. Two examples are given, illustrating realistic problems in experimental design in molecular phylogenetics and suggesting more general conclusions about the choice of genomic regions, sequence lengths and taxa for evolutionary studies.