The paper introduces UFBoot2, an advanced version of the ultrafast bootstrap approximation (UFBoot) method, which significantly improves the speed and accuracy of phylogenetic bootstrap analyses. UFBoot2 addresses the limitations of standard bootstrap (SBS) and rapid bootstrap (RBS) methods by reducing computational time and minimizing the risk of overestimating branch supports due to polytomies or severe model violations. The key improvements include a modified pruning algorithm that leverages the reversible property of sequence evolution models, SIMD optimization for concurrent likelihood computation, and a technique to correct for overestimation in polytomies. UFBoot2 also provides several resampling strategies for phylogenomic data, including MSA-site resampling, gene resampling, and gene-site resampling. Benchmark tests show that UFBoot2 is 778 times faster than SBS and 8.4 times faster than RBS on tested datasets. The method is implemented in the IQ-TREE software package and is freely available. The authors recommend using UFBoot2 for unbiased bootstrap support estimation under mild model violations and UFBoot2 + NNI for severe model violations.The paper introduces UFBoot2, an advanced version of the ultrafast bootstrap approximation (UFBoot) method, which significantly improves the speed and accuracy of phylogenetic bootstrap analyses. UFBoot2 addresses the limitations of standard bootstrap (SBS) and rapid bootstrap (RBS) methods by reducing computational time and minimizing the risk of overestimating branch supports due to polytomies or severe model violations. The key improvements include a modified pruning algorithm that leverages the reversible property of sequence evolution models, SIMD optimization for concurrent likelihood computation, and a technique to correct for overestimation in polytomies. UFBoot2 also provides several resampling strategies for phylogenomic data, including MSA-site resampling, gene resampling, and gene-site resampling. Benchmark tests show that UFBoot2 is 778 times faster than SBS and 8.4 times faster than RBS on tested datasets. The method is implemented in the IQ-TREE software package and is freely available. The authors recommend using UFBoot2 for unbiased bootstrap support estimation under mild model violations and UFBoot2 + NNI for severe model violations.