In three experiments, reaction times for same-different judgments were obtained for pairs of words, pronounceable nonwords (pseudcwords), and unpronounceable nonwords. The stimulus strings were printed either in a single letter case or in one of several mixtures of upper-and lowercase letters. In Experiment 1, the stimuli were common one-and two-syllable words; in Experiment 2, the stimuli included both words and pseudowords; and in Experiment 3, words, pseudowords, and nonwords were used. The functional visual units for each string type were inferred from the effects that the number and placement of letter case transitions had on same reaction time judgments. The evidence indicated a preference to encode strings in terms of multiletter perceptual units if they are present in the string. The data also suggested that whole words can be used as functional visual units, although the ex~t of their use depends on contextual parameters such as knowledge that a word will be presented"For over a century, the question of how we perceive letter strings, and words in particular, has both fascinated and troubled experimental psychologists. For some, it has seemed obvious that we perceive individual letters and assemble a word from them. Others have maintained that we perceive whole words directly, while still others have postulated intermediate perceptual units such as spelling patterns, syllables, and vocalic center groups (Smith & Spoehr, 1974). To the casual observer it must seem perplexing that psychologists have been unable to resolve such a seemingly simple issue in favor of one or another alternative. However, many investigations have resulted in conflicting indications of what constitutes the appropriate visual unit in word perception.It is understandable that there is reluctance to abandon the attempt to identify a single perceptual unit. If more than one perceptual unit is hypothesized, then an adequate model must specify which units exist and how these units are coordinated. The model to which we subscribe is a process model similar to the one proposed by LaBerge and Samuels