In this article the suitability of two different tree representations for t
he construction of phylogenetic trees with genetic algorithms is examined.
On the one hand tree topologies are represented by means of a distance matr
ix while on the other hand the Prufer number tree representation is used. T
o assess the adequacy of both approaches a set of recently proposed quality
criteria is used. The quality criteria can be used to monitor genetic algo
rithm approaches differing in configuration and setup, fitness function, or
representation. In addition to the criteria for the repeatability of the o
ptimization, criteria for the coverage of the search space are also used. O
n the basis of the optimization results of simulated data, the quality crit
eria show, in contrast to the error plots, a clear difference in the effici
ency of both representations. It is concluded that the Prufer number repres
entation yields a diverse set of good quality topologies while the distance
matrix representation mainly returns the optimal topology. (C) 1999 John W
iley & Sons, Inc.