Earlier research has found that phonological markedness constraints (for example, against stress clash or sibilant sequences) statistically influence speakers’ choices between particular syntactic constructions and between synonymous words. In this study, we test phonological constraints not just in particular cases, but across the board. We employ a novel method that statistically models the distribution of word bigrams (consecutive two-word sequences) and how this distribution is influenced by phonological constraints. Our study of multiple corpora indicates that several phonological constraints do indeed play a statistically significant role in English sentence formation. We also show that by examining particular subsets of the corpora we can diagnose the mechanisms whereby phonologically marked sequences come to be underrepresented. We conclude by discussing modes of grammatical organization compatible with our findings.