RAxML-III is a fast program for maximum likelihood-based inference of large phylogenetic trees. It allows computation of 1,000-taxon trees in less than 24 hours on a single PC processor. The program was compared to PHYML and MrBayes, the fastest implementations for maximum likelihood and Bayesian inference. While RAxML-III performs worse than PHYML and MrBayes on synthetic data, it clearly outperforms both programs on real data alignments in terms of speed and final likelihood values. RAxML-III incorporates the HKY85 and GTR models of DNA sequence evolution and can optimize all free model parameters. It is able to perform a maximum likelihood estimate of per-site evolutionary rates for HKY85. The program is faster and more accurate than PHYML and MrBayes on real data. RAxML-III is also more efficient in terms of memory usage compared to MrBayes and PHYML. However, it provides less modeling flexibility than MrBayes and PHYML, as it does not handle protein sequence data or estimate the proportion of invariable sites or the Γ model of rate heterogeneity. RAxML-III is designed to allow for the implementation of relatively coarse-grained distributed and parallel versions, which do not rely on expensive hardware platforms. The program is intended to serve as a useful tool to improve Bayesian inference and vice versa. The results show that RAxML-III and PHYML represent very fast and accurate conventional maximum likelihood programs, which allow for sequential inference of large trees within reasonable times on standard PC architectures.RAxML-III is a fast program for maximum likelihood-based inference of large phylogenetic trees. It allows computation of 1,000-taxon trees in less than 24 hours on a single PC processor. The program was compared to PHYML and MrBayes, the fastest implementations for maximum likelihood and Bayesian inference. While RAxML-III performs worse than PHYML and MrBayes on synthetic data, it clearly outperforms both programs on real data alignments in terms of speed and final likelihood values. RAxML-III incorporates the HKY85 and GTR models of DNA sequence evolution and can optimize all free model parameters. It is able to perform a maximum likelihood estimate of per-site evolutionary rates for HKY85. The program is faster and more accurate than PHYML and MrBayes on real data. RAxML-III is also more efficient in terms of memory usage compared to MrBayes and PHYML. However, it provides less modeling flexibility than MrBayes and PHYML, as it does not handle protein sequence data or estimate the proportion of invariable sites or the Γ model of rate heterogeneity. RAxML-III is designed to allow for the implementation of relatively coarse-grained distributed and parallel versions, which do not rely on expensive hardware platforms. The program is intended to serve as a useful tool to improve Bayesian inference and vice versa. The results show that RAxML-III and PHYML represent very fast and accurate conventional maximum likelihood programs, which allow for sequential inference of large trees within reasonable times on standard PC architectures.