The bootstrap is a statistical technique that is widely used to assess confidence limits on phylogenies. We show that the power of the bootstrap test is lower than those of the C and S tests suggested by Felsenstein, unless the critical value employed in the bootstrap test is correctly selected. If the 95% critical value is used, the bootstrap proportions are underestimates of the confidence level when the number of possible alternative topologies is three or more; the degree of underestimation increases with the number of competing alternative topologies. To overcome this problem, we propose the complete-and-partial bootstrap technique as a method for obtaining an unbiased estimate of the confidence level. The method is based on a multinomial model of many alternatives among which the choice is to be made. The complete-and-partial bootstrap technique can be used to estimate the effective number of competing alternative topologies and the confidence level of the monophyly of a particular group of taxa or of an inferred tree topology. This approach can be used with the maximum parsimony or neighbor-joining tree reconstruction method.