Students in grades 2, 4, and 6 were tested in two blocks of visual comparison trials. In one block, the primary task was to decide whether two words matched, and in the other block the task was to decide whether a single letter target matched the first letter of a word. On some trials in both blocks, the word comparison item was omitted and subjects were to decide whether a “7” occurred in the replacement display. On those probe trials, the “7” occurred either to the left or right of the display. In all grades, latencies for the primary task were faster for whole-word than for first-letter decisions. On the probe trials, latencies were faster when the probe item appeared in the first position of the display, but only in the block of first-letter trials. Moreover, the first-position advantage on probe trials was greater for good than for poor readers. The results were interpreted to be consistent with the hypothesis that both good and poor readers process words holistically. Poor readers are not more likely than good readers to attend to single-letter units in visual memory (cf. LaBerge & Samuels, 1974). Rather, poor readers are deficient in analyzing a word's components (cf. Wolford & Fowler, 1984).