T',e 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 the association was playing a small role, but that recognition performance on pairs of words was highly predictable from frequency information. 'How-