Murdock (1974) located the difference in recognition between common and rare words in the lower frequency of false positives to rare words. Morris & Reid (1974) have similarly located the superiority in recognition of easy to image (high I) words in their lower false positive rate. However, Gregg's (1976) explanation of the word frequency effect in terms of encoding variability and the distinctiveness of word attributes predicts more hits as well as fewer false positives to rare words. Gregg's prediction was confirmed by the experiment reported below. A subsequent survey of dictionary definitions of words differing in frequency and I-value further supported Gregg's model and was incompatible with that of Glanzer & Bowles (1976). A model based upon the likelihood of encoding variability and the distinctiveness of the attributes defining each word meaning accounts for the differences in recognition for items varying in I-value and frequency, and explains Galbraith & Underwood's (1973) finding that abstract words are perceived as more common than concrete words.