Two experiments in which 12-year old children had to spell Dutch inflected verb forms are reported. Both experiments focus on homophone dominance, i.e., the fact that spellers tend to make more intrusion errors on the lower-frequency form than on the higher-frequency one. Homophone-induced errors are studied at the level of full forms in Experiment 1 and at the sublexical level in Experiment 2. In Experiment 1 the children had to fill out two types of verb forms with homophones in their 1st (verb-final d) and 3rd person (verb-final dt) singular present tense. In both types the two verb forms had a very low frequency but the 1st person homophone of one type had the same spelling as a noun or adjective ending in d, or ended in such a word. The children made significantly more d intrusions on the 3rd person of these verbs than on the 3rd person of control verbs. In Experiment 2 three types of past tenses with stem-final d and suffix de had to be filled out, differing in the type of orthographic cluster preceding the stem-d. The pattern of results supports an account in which phonologically similar verbs are activated by the sublexical word-final sound sequence, which gives rise to intrusion errors when that sequence is homophonous between past tenses ending in de and dde. As the same phenomenon manifests itself at the levels of full forms and sublexical patterns, a model that automatically captures systematic correspondences between phonological and orthographic representations can best explain these findings. Connectionist and exemplar accounts are the prime candidates for such an explanation. Words-and-rules models, on the other hand, have several problems explaining the data.