What is it that we know when we know the spellings of words? This chapter reviews current understanding of the answer to that question, focusing on evidence from written word production (spelling) and briefly reviewing convergent evidence from reading. We first establish that orthographic knowledge is independent from spoken word knowledge. With regard to the nature of orthographic representations, evidence indicates that orthographic knowledge is represented in a modality-independent code organized into units corresponding to morphemes, digraphs, and letters. Furthermore, a detailed examination of the representation of letter units reveals that they are multidimensional feature bundles specifying letter identity, consonant-vowel status, syllabic role, and letter position. This chapter shows that to know the spellings of words is to learn and process orthographic representations that are abstract, complex, and richly structured mental objects.