Background. Antibiotic duration is often longer than necessary. Understanding the reasons for variability in antibiotic duration can inform interventions to reduce prolonged antibiotic use. We aim to describe patterns of interphysician variability in prescribed antibiotic treatment durations and determine physician predictors of prolonged antibiotic duration in the community setting. Methods. We performed a retrospective cohort analysis of family physicians in Ontario, Canada, between 1 March 2016 and 28 February 2017, using the Xponent dataset from IQVIA. The primary outcome was proportion of prolonged antibiotic course prescribed, defined as >8 days of therapy. We used multivariable logistic regression models, with generalized estimating equations to account for physician-level clustering to evaluate predictors of prolonged antibiotic courses. Results. There were 10 616 family physicians included in the study, prescribing 5.6 million antibiotic courses. There was substantial interphysician variability in the proportion of prolonged antibiotic courses (median, 33.3%; interdecile range, 13.5%-60.3%). In the multivariable regression model, later physician career stage, rural location, and a larger pediatric practice were significantly associated with greater use of prolonged courses. Prolonged courses were more likely to be prescribed by late-career physicians (adjusted odds ratio [aOR], 1.48; 95% confidence interval, 1.38-1.58) and mid-career physicians (aOR, 1.25; 1.16-1.34) when compared to early-career physicians. Conclusions. We observed substantial variability in prescribed antibiotic duration across family physicians, with durations particularly long among late-career physicians. These findings highlight opportunities for community antimicrobial stewardship interventions to improve antibiotic use by addressing practice differences in later-career physicians.