Starting from a generic architecture for reading words in alphabetic scripts, we examine the special status of letters as the building block of single word reading. After briefly describing the overall architecture that defines the interaction between orthographic and phonological processes during silent reading for meaning, we then focus on orthographic processing. We describe the nature of orthographic representations as hypothesized in our approach and we discuss how such representations might be learned during reading acquisition. We present the hypothesis that such learning involves the adaptation of basic object identification mechanisms to the specific constraints of reading, and we provide examples of this adaptation. In the light of this, we then compare the function of letters as constituents of written words relative to the role of object parts in other kinds of familiar visual stimuli (e.g. faces, numbers). We explain why we think letters must have a special status and we provide some preliminary empirical evidence in favor of this special status for letters as parts of words. Keywords: reading; orthography; visual word recognition; orthographic learning; letter strings; object identification