Educational gradients in parental leave length are opposite for women and men: highly educated women return to work faster than those with low education while highly educated men are absent longer than less educated men. Explanations for the opposite gradients are typically made at the individual- or couple-level. To date, no quantitative study has documented whether the opposite educational gradients hold also within workplaces. In this study, we use employer-employee matched Swedish register data with fixed-effects models to examine whether the educational gradient applies also among co-workers in the same workplace. The results show that three-quarters of the educational effect typically attributed to the individual father disappeared when comparing fathers within workplaces. The educational gradient of mothers remained largely unchanged. These findings provide the first population-level evidence for the primacy of the workplace in determining fathers' care choices.