A growing body of evidence supports a multi-level view of visual word recognition incorporating representations at an intermediate level where multiletter units can be activated directly by supraletter features (e.g. Drewnowski & Healy, 1977). In the present study, priming in an alphabetic decision task was exploited to investigate the existence of directly activated multiletter units. Subjects were required to make a discrimination response to test stimuli that could be targets or foils. Targets were single letters or consonant-bigrams that were present or absent in an immediately preceding word, and foils were single keyboard characters or a character plus a letter. Experiment I verified an earlier finding that responses to consonantbigrams are facilitated when they appear in a prime word, whereas responses to the constituent letters of those bigrams are not facilitated (Greenberg & Vellutino, 1988). In addition, responses to primed bigrams were faster than responses to primed single letters. Experiments 2 to 4 revealed that bigram priming in the absence of letter priming occurs only when both primes and targets appear in lower-case type. It is concluded that separable multiletter units are represented in the visual word recognition system and are directly activated by supraletter visual features. The findings are discussed in the context of current theories of visual word recognition.