Abstract Canonical correlation analysis, as described by Webb and Bryson, Quaternary Research 1972, provides a means of reconstructing past climates quantitatively from fossil pollen using a pollen-climate transfer function. This paper presents a method for analysis of variance of the transfer function model. This method is used to identify ecological relationships among the pollen and climate variables, to select climatically sensitive taxa, and to investigate the importance of site factors. Several criteria are presented, in addition to those used by Webb and Bryson, for choosing canonical variate pairs to include in the transfer function model, namely: the variate pair relationships should be ecologically meaningful; the transfer function model should yield stable paleoclimatic estimates; and, the variate pair relationships should be statistically meaningful. The application of these criteria to the set of variate pairs used in the transfer function model of Webb and Bryson is described and modifications are suggested.