The article addresses the issue of whether disabled readers can be instructed to use within-word (multiletter) units. Several analogy methods were employed. In Experiments 1 and 2 short lists of words that differ in only one letter were repeatedly practiced. In Experiment 3 the subjects blended written and spoken words in which the place of segmentation was systematically manipulated. Subjects were tested in all experiments on the reading of practiced words, on nonpracticed words that orthographically as well as phonologically resemble the practiced words, and on nonpracticed nonsimilar words. The findings indicate that: (1) All types of practice were beneficial for the recognition of the practiced words. (2) Transfer to the reading of novel words appeared only when graphemic and orthographic aspects of words were strongly emphasized. (3) Transfer effects were based upon the improved handling of multiletter within-word units when word lists with auditory prompts were practiced, but on improved (single) letter-sound conversions when the analysis and synthesis of isolated words and word parts was trained. (4) At least in Dutch, segmentations according to the onset-rime principle were not more useful than segmenting written and spoken words at other boundaries.