Quantitative constraint-based theories of optionality typically aim to model the frequency with which an individual speaker’s grammar maps one input onto various output forms. But existing studies in this area use population-level frequency data as a proxy for the frequencies produced by an individual speaker’s grammar. This practice is problematic for three reasons. First, population-level variation could result from aggregation over individuals with differing categorical behavior, not individual variation at all. Second, some theories predict that individuals with identical relative constraint rankings/weights can produce those outputs with different frequencies, while others predict uniform frequencies across speakers; these prediction cannot be tested without individual-level data. Third, if indeed individuals differ in their output frequencies, population-level averages may disguise true patterns of individual variation. This paper addresses these shortcomings through a corpus study of optional schwa deletion in French. We find that individual speakers do in fact show variable omission of schwa. We also find variation between speakers in the frequency of schwa’s appearance. Taken together, these results confirm that for French schwa, formal theories are correct to ascribe variation to a single speaker’s grammar, and they favor theories that decouple the means of producing variation from the account of output frequency.