The paper introduces UFBoot, an ultrafast approximation method for phylogenetic bootstrap, which significantly reduces computational time compared to traditional methods like RAxML rapid bootstrap (RBS). UFBoot combines the resampling estimated log-likelihood method with a collection scheme of candidate trees, using the IQPNNI algorithm to sample trees efficiently. A stopping rule based on the correlation coefficient of split support values ensures convergence. Extensive simulations show that UFBoot provides unbiased support values and is robust against moderate model violations. The method is implemented in the IQ-TREE software package, which allows users to reconstruct ML trees, bootstrap trees, and consensus trees with support values in a single run. UFBoot outperforms RBS in terms of computational time and accuracy, making it a valuable tool for large-scale phylogenetic analysis.The paper introduces UFBoot, an ultrafast approximation method for phylogenetic bootstrap, which significantly reduces computational time compared to traditional methods like RAxML rapid bootstrap (RBS). UFBoot combines the resampling estimated log-likelihood method with a collection scheme of candidate trees, using the IQPNNI algorithm to sample trees efficiently. A stopping rule based on the correlation coefficient of split support values ensures convergence. Extensive simulations show that UFBoot provides unbiased support values and is robust against moderate model violations. The method is implemented in the IQ-TREE software package, which allows users to reconstruct ML trees, bootstrap trees, and consensus trees with support values in a single run. UFBoot outperforms RBS in terms of computational time and accuracy, making it a valuable tool for large-scale phylogenetic analysis.