The influence of a patient's quantitative skills (numeracy) on the management of diabetes is only partially understood. To examine the association between diabetes-related numeracy and glycemic control and other diabetes measurements. Cross-sectional survey. 2 primary care and 2 diabetes clinics at 3 medical centers. 398 adult patients with type 1 or type 2 diabetes mellitus enrolled between March 2004 and November 2005. Health literacy, general numeracy, and diabetes-related numeracy assessed by using the Rapid Estimate of Adult Literacy in Medicine; the Wide Range Achievement Test, 3rd edition; and the Diabetes Numeracy Test (DNT), respectively. The primary outcome was most recent level of hemoglobin A1c. Additional measurements were diabetes knowledge, perceived self-efficacy of diabetes self-management, and self-management behaviors. The median DNT score was 65% (interquartile range, 42% to 81%). Common errors included misinterpreting glucose meter readings and miscalculating carbohydrate intake and medication dosages. Lower DNT scores were associated with older age, nonwhite race, fewer years of education, lower reported income, lower literacy and general numeracy skills, lower perceived self-efficacy, and selected self-management behaviors. Patients scoring in the lowest DNT quartile (score <42%) had a median hemoglobin A1c level of 7.6% (interquartile range, 6.5% to 9.0%) compared with 7.1% (interquartile range, 6.3% to 8.1%) in those scoring in the highest quartile (P = 0.119 for trend). A regression analysis adjusted for age, sex, race, income, and other factors found a modest association between DNT score and hemoglobin A1c level. Causality cannot be determined in this cross-sectional study, especially with its risk for unmeasured confounding variables. Poor numeracy skills were common in patients with diabetes. Low diabetes-related numeracy skills were associated with worse perceived self-efficacy, fewer self-management behaviors, and possibly poorer glycemic control.