Background: Guidelines recommend that relatives be present during cardiopulmonary resuscitation (CPR). This randomised trial investigated the effects of two different behaviour patterns of relatives on rescuers’ perceived stress and quality of CPR. Material and methods: Teams of three to four physicians were randomised to perform CPR in the presence of no relatives (control group), a withdrawn relative, or an agitated relative, played by actors according to a scripted role, and to three different models of leadership (randomly determined or left open by the team or tutor). The scenarios were video-recorded. Hands-on time was primary, and the secondary outcomes comprised compliance to CPR algorithms, perceived workload, and the influence of leadership. Results: 1229 physicians randomised to 366 teams took part. The presence of a relative did not effect hands-on time (91% [87–93] vs. 92% [88–94] for “withdrawn” and 92 [88–93] for “agitated” relatives; p = 0.15). The teams interacted significantly less with a “withdrawn” than with an “agitated” relative (11 [7–16]% vs. 23 [15–30]% of the time spent for resuscitation, p < 0.01). The teams confronted with an “agitated” relative showed more unsafe defibrillations, higher ventilation rates, and a delay in starting CPR (all p < 0.05 vs. control). The presence of a relative increased frustration, effort, and perceived temporal demands (all <0.05 compared to control); in addition, an “agitated” relative increased mental demands and total task load (both p < 0.05 compared to “withdrawn” and control group). The type of leadership condition did not show any effects. Conclusions: Interaction with a relative accounted for up to 25% of resuscitation time. Whereas the presence of a relative per se increased the task load in different domains, only the presence of an “agitated” relative had a marginal detrimental effect on CPR quality (GERMAN study registers number DRKS00024761).