Detection of natural selection operating at the amino acid sequence level is important in the study of molecular evolution. Single-site analysis and one-dimensional window analysis can be used to detect selection when the biological functions of amino acid sites are unknown. Single-site analysis is useful when selection operates more or less constantly over evolutionary time, but less so when selection operates temporarily. One-dimensional window analysis is more sensitive than single-site analysis when the functions of amino acid sites in close proximity in the linear sequence are similar, although this is not always the case. Here I present a three-dimensional window analysis method for detecting selection given the three-dimensional structure of the protein of interest. In the three-dimensional structure, the window is defined as the sphere centered on the alpha-carbon of an amino acid site. The window size is the radius of the sphere. The sites whose alpha-carbons are included in the window are grouped for the neutrality test. The window is moved within the three-dimensional structure by sequentially moving the central site along the primary amino acid sequence. To detect positive selection, it may also be useful to group the surface-exposed sites in the window separately. Three-dimensional window analysis appears not only to be more sensitive than single-site analysis and one-dimensional window analysis but also to provide similar specificity for inferring positive selection in the analyses of the hemagglutinin and neuraminidase genes of human influenza A viruses. This method, however, may fail to detect selection when it operates only on a particular site, in which case single-site analysis may be preferred, although a large number of sequences is required.