The paper introduces RAxML-III, a program designed for rapid maximum likelihood-based inference of large phylogenetic trees. The authors highlight the computational challenges of using statistical models like maximum likelihood or Bayesian inference for large trees, which are typically limited to around 100 taxa due to their high complexity. RAxML-III aims to address this by enabling the computation of 1,000-taxon trees in less than 24 hours on a single PC processor. The program is compared to other fast implementations such as PHYML and MrBayes, showing superior performance on real data alignments in terms of speed and final likelihood values. The introduction discusses the limitations of traditional methods and the benefits of RAxML-III, including its ability to explore a larger search space and generate multiple starting trees. The algorithm's implementation and performance on various datasets are detailed, demonstrating its effectiveness in both simulated and real-world data. The paper also provides a comprehensive benchmark set for future researchers to assess the performance of maximum likelihood programs.The paper introduces RAxML-III, a program designed for rapid maximum likelihood-based inference of large phylogenetic trees. The authors highlight the computational challenges of using statistical models like maximum likelihood or Bayesian inference for large trees, which are typically limited to around 100 taxa due to their high complexity. RAxML-III aims to address this by enabling the computation of 1,000-taxon trees in less than 24 hours on a single PC processor. The program is compared to other fast implementations such as PHYML and MrBayes, showing superior performance on real data alignments in terms of speed and final likelihood values. The introduction discusses the limitations of traditional methods and the benefits of RAxML-III, including its ability to explore a larger search space and generate multiple starting trees. The algorithm's implementation and performance on various datasets are detailed, demonstrating its effectiveness in both simulated and real-world data. The paper also provides a comprehensive benchmark set for future researchers to assess the performance of maximum likelihood programs.