We propose an order index, φ, which quantifies the notion of "life at the edge of chaos" when applied to genome sequences. It maps genomes to a number from 0 (random and of infinite length) to 1 (fully ordered) and applies regardless of sequence length. The 786 complete genomic sequences in GenBank were found to have φ values in a very narrow range, 0.037±0.027. We show this implies that genomes are halfway towards being completely random, namely, at the edge of chaos. We argue that this narrow range represents the neighborhood of a fixed-point in the space of sequences, and genomes are driven there by the dynamics of a robust, predominantly neutral evolution process.The Edge of chaos originally refers to the state of a computational system, such as cellular automata, when it is close to a transition to chaos, and gains the ability for complex information processing [1,2,3]. The notion has since been used to describe biological states, and life in general, on the assumption that life necessarily involves complex computation [4]. In model systems such as cellular automata, there are well defined procedures for recognizing the change in computational cab ability during the transition from non-chaotic to chaotic states [1,3]. However, these have not been adapted to the wider biological context, even for the simplest of organisms. But if we represent a living organism by its genome, view evolution as a dynamical process that drives genomes in the space of sequences, and consider chaos as a state of genome randomness, then we have a framework within which the meaning of "life occurs at the edge of chaos" may be investigated. Genomes, linear sequences written in the four chemical letters, or bases, A (adenine), C (cytosine), G (guanine) and T (thymine) and often referred to as books of life, regulate the functioning of organisms through the many kinds of codes embedded in them (there are also non-textual post-translational regulations; see, e.g. [5]). When genomes are seen as texts, they have several key properties reflecting their complexity, including long-range correlations and scale invariance [6,7,8] (although this topic is debated [9]), self-similarity [10,11,12,13], and distinctive Shannon redundancy [14,15,16]. However, these properties do not give a measure of the proximity of a genome to chaos or randomness. Before the edge-of-chaos notion can be explored, one needs to have a quantity that measures the randomness of genomes as texts.Here we analyze genomes in terms of the frequency of occurrence of k-letter words, called k-mers, where k is a small integer [17]. For a given k, the 4 k types of k-mers are partitioned into k+1 "m-sets", m=0-k. An m-set is composed of all the k-mers containing m and only m A or T's. There are τ m =2 k " k m « types of k-mers in an m-set.The reason for partitioning the k-mers according to ATcontent for statistical purposes is that although the A:T and C:G ratios are invariably close to 1, [18,19,20], the AT to GC ratio may differ significantly. This partition is needed for prevent...