The purpose of the eight experiments was to assess the role which associations between two words played in recognition decisions. The evidence on weak associations established in the laboratory indicated that association was playing a small role, but that the recognition performance on pairs of words was highly predictable from frequency information. However, the use of strongly associated words indicated that the strength of the association per se was not a critical variable in recognition performance. A post hoc expansion of frequency theory was proposed. Some unexpected findings included criterion differences in making frequency judgments as compared with recognition decisions, and criterion differences in recognition tests on homonym pairs as compared with other classes of word pairs. (Author)