The purpose of the experiments was to determine the automatic use of large or small word reading units in young readers in the absence of word decoding strategies. Picture-word Stroop interference was examined from four types of conflicting labels: (a) words containing both highly predictable grapheme-phoneme correspondence (GPC) units and highly consistent rime units (henceforth, Hi-GPC + Hi-Rime); (b) words with highly predictable GPC units and less consistent rime units (Hi-GPC + Lo-Rime); (c) words with low predictability GPC units and highly consistent rime units (Low GPC + High Rime); (d) nonwords that contained both highly predictable GPC and highly consistent rime units. Naming time for pictures containing these labels was compared against that for pictures with random letter strings or no labels. In Experiment 1, Stroop interference was examined in first, second, and third grade children to determine whether there was developmental change in the presence of rime or GPC interference. In Experiment 2, Stroop interference was examined as a function of relative reading skill in first grade children. In Experiment 3, Stroop interference in adults was compared to the use of rime or GPC pronunciation strategies for nonword reading. In all experiments, Stroop interference in picture naming was longer for pictures with highly predictable GPC unit labels than less predictable GPC unit labels. However, in Experiment 3, even though adults showed interference from predictable GPC units in the Stroop task, they always preferred rime pronunciation for ambiguous non-words in the nonword reading task. It is argued that the current experiments provide evidence for a flexible units model.