To diagnose faults in a process plant, teams of physically separated operators must work together. They have access to differing information sources and need to rely on verbal communication, which can lead to various problems. Shared mental models facilitate team interaction, but most previous research collected information about mental models post-hoc. The present study manipulated the similarity of mental models by varying the amount of knowledge that participants received in an instruction phase. Three groups of ten dyads jointly diagnosed five faults in a simulated process plant, while either sharing a task model, a team model, or both. Teams that shared both models communicated less and used communication categories that differed from teams who only shared one model. Performance did not differ between groups, except for fewer rule violations when sharing both models. The findings suggest it may be beneficial to train all team members on the complete team task.