In the empathic accuracy paradigm, perceivers make inferences about the naturalistically occurring thoughts and feelings of stimulus persons, and these inferences are scored for accuracy against the stimulus persons' self-reported thoughts and feelings. The present study investigated sources of accuracy in this paradigm by presenting the stimulus tape in several cue modalities (full video, audio, transcript, or silent video) and with differing instructions (infer thoughts and feelings, infer thoughts, or infer feelings). Verbal information contributed the most to accuracy, followed by vocal nonverbal cues. Visual nonverbal cues contributed the least, though still at levels above zero. When asked to infer feelings, perceivers appeared to shift attention toward visual nonverbal cues and away from verbal cues, and the reverse occurred when they were asked to infer thoughts. The study contributes to understanding of factors contributing to accuracy in the empathic accuracy paradigm.