CONSEL is a program for assessing the confidence of phylogenetic tree selection by calculating p-values for the trees. It primarily calculates the p-value of the Approximately Unbiased (AU) test using a multi-scale bootstrap technique, which is less biased than other conventional p-values such as Bootstrap Probability (BP), Kishino–Hasegawa (KH), Shimodaira–Hasegawa (SH), and Weighted Shimodaira–Hasegawa (WSH) tests. CONSEL can calculate these p-values from the output of phylogeny programs like Molphy, PAML, and PAUP*. It is applicable to various problems where BPs are available. The program consists of two parts: makermt and consel. Makermt reads the matrix of site-wise log-likelihoods of candidate trees and generates bootstrap replicates using the RELL resampling method. Consel reads these replicates and calculates several p-values for the candidate trees, including BP, KH, SH, WSH, and AU tests. The AU test, developed to reduce test bias, is based on the same theory as Efron et al. (1996) but with higher accuracy and simpler implementation. It assumes the smoothness of hypothesis region boundaries, though it may fail in some cases, in which case SH or WSH tests are used. The AU test calculates p-values using a multi-scale bootstrap approach, generating multiple sets of bootstrap replicates with varying sample sizes. The p-values are derived from the change in the ratio of counts to replicates across different scales. CONSEL accepts three types of input files: rmt, rep, and cnt. The rmt file contains bootstrap replicates of log-likelihoods, the rep file contains test statistics, and the cnt file contains counts for all candidate trees. CONSEL is applicable to a wide range of problems where BPs are available, making it a versatile tool for phylogenetic analysis. It supports various criteria beyond likelihood, and the generality of its input files allows it to be used for different types of hypothesis testing. The program is supported by grants from JSPS and MEXT.CONSEL is a program for assessing the confidence of phylogenetic tree selection by calculating p-values for the trees. It primarily calculates the p-value of the Approximately Unbiased (AU) test using a multi-scale bootstrap technique, which is less biased than other conventional p-values such as Bootstrap Probability (BP), Kishino–Hasegawa (KH), Shimodaira–Hasegawa (SH), and Weighted Shimodaira–Hasegawa (WSH) tests. CONSEL can calculate these p-values from the output of phylogeny programs like Molphy, PAML, and PAUP*. It is applicable to various problems where BPs are available. The program consists of two parts: makermt and consel. Makermt reads the matrix of site-wise log-likelihoods of candidate trees and generates bootstrap replicates using the RELL resampling method. Consel reads these replicates and calculates several p-values for the candidate trees, including BP, KH, SH, WSH, and AU tests. The AU test, developed to reduce test bias, is based on the same theory as Efron et al. (1996) but with higher accuracy and simpler implementation. It assumes the smoothness of hypothesis region boundaries, though it may fail in some cases, in which case SH or WSH tests are used. The AU test calculates p-values using a multi-scale bootstrap approach, generating multiple sets of bootstrap replicates with varying sample sizes. The p-values are derived from the change in the ratio of counts to replicates across different scales. CONSEL accepts three types of input files: rmt, rep, and cnt. The rmt file contains bootstrap replicates of log-likelihoods, the rep file contains test statistics, and the cnt file contains counts for all candidate trees. CONSEL is applicable to a wide range of problems where BPs are available, making it a versatile tool for phylogenetic analysis. It supports various criteria beyond likelihood, and the generality of its input files allows it to be used for different types of hypothesis testing. The program is supported by grants from JSPS and MEXT.