In conversational speech, it is very common for words' segments to be reduced or deleted. However, previous research has consistently shown that during spoken word recognition, listeners prefer words' canonical pronunciation over their reduced pronunciations (e.g., pretty pronounced [word omitted] vs. [word omitted]), even when the latter are far more frequent. This surprising effect violates most current accounts of spoken word recognition. The current study tests the possibility that words' orthography may be 1 factor driving the advantage for canonical pronunciations during spoken word recognition. Participants learned new words presented in their reduced pronunciation (e.g., [word omitted]), paired with 1 of 3 spelling possibilities: (a) no accompanying spelling, (b) a spelling consistent with the reduced pronunciation (a reduced spelling, e.g., "troddy"), or (c) a spelling consistent with the canonical pronunciation (a canonical spelling, e.g., "trotty"). When listeners were presented with the new words' canonical forms for the first time, they erroneously accepted them at a higher rate if the words had been learned with a canonical spelling. These results remained robust after a delay period of 48 hr, and after additional learning trials. Our findings suggest that orthography plays an important role in the recognition of spoken words and that it is a significant factor driving the canonical pronunciation advantage observed previously.