To clarify the link between anomalous letter processing and developmental dyslexia, we examined the impact of surrounding contours on letter vs. pseudo-letter processing by three groups of children--phonological dyslexics and two controls, one matched for chronological age, the other for reading level--and three groups of adults differing by schooling and literacy--unschooled illiterates and ex-illiterates, and schooled literates. For pseudo-letters, all groups showed "congruence effects" (CE: better performance for targets surrounded by a congruent than by an incongruent shape). In contrast, for letters, only dyslexics exhibited a CE, strongly related to their phonological recoding abilities even after partialling out working memory, whereas the reverse held true for the pseudo-letter CE. In illiterate adults, the higher letter knowledge, the smaller their letter CE; their letter processing was immune (to some extent) to inference from surrounding information. The absence of a letter CE in illiterates and the positive CE in dyslexics have their origin in different aspects of the same ability, i.e. phonological recoding.