In the absence of bias between the two DNA strands for mutation and selection, the base composition within each strand should be such that A=T and C=G (this state is called Parity Rule type 2, PR2). At a genome scale, i.e. when considering the base composition of a whole genome, PR2 is a good approximation, but there are local and systematic deviations. The question is whether these deviations are a consequence of an underlying bias in mutation or selection. We have tried to review published hypotheses to classify them within the mutational or selective group. This dichotomy is, however, too crude because there is at least one hypothesis based simultaneously upon mutation and selection. (C) 1999 Elsevier Science B.V. All rights reserved.