This study investigates schwa deletion in American English using the Buckeye Corpus. To determine the contributions of grammatical (e.g., sonority of onset clusters) and non-grammatical (e.g., lexical frequency) factors to schwa deletion, the distribution of schwa-deleted word forms is examined in terms of phonological, morphological, and lexical environments. As expected, schwa deletion occurred more frequently in post-stressed environments than in pre-stressed environments. Of all the factors, the most significant factor affecting deletion rates in the post-stressed environment was the sonority profile of onset clusters that arise after deletion: deletion rates are higher when the prospective consonant cluster forms a sonority rise (large rise and small rise) than when it forms a sonority fall. The results show that the Sonority Sequencing Principle alone cannot explain the observed patterns of variable schwa deletion. This paper suggests that schwa deletion rates are modulated by English speakers’ knowledge of the universal sonority preferences. The nature of this knowledge regardine sonority preferences will be further discussed.