Given recent calls for enhancing multilevel leadership research, the present study uses event-level data to test the proposition that follower leadership perceptions result from aggregated personal observations of leader behavior. Although this proposition implies causality, it has been mostly tested with between-person retrospective data, coupled with correlational study designs. Our data refer to the perceived nature of leadership messages communicated during the most recent conversation with followers. These event-level data were used in a feedback intervention coupled with goal setting and population norms devised to change leader verbal behavior. The study was designed as a randomized field experiment in a manufacturing company in which supervisors in the experimental group received two feedback sessions regarding the extent to which conversation participants perceived their messages as transformative, transactive (corrective), or passive. Supervisors in the control group received no feedback. The data indicated significant changes in verbal leader behavior for experimental group supervisors, remaining unchanged in the control group. Such changes resulted in matching changes in follower leadership perceptions, measured 8 weeks before and after the intervention. Furthermore, performance outcomes improved in the experimental group, remaining unchanged in the control group. Implications for leadership research are discussed.