The ability to infer the emotional states of others is central to our everyday interactions. These inferences can be drawn from several different sources of information occurring simultaneously in the communication situation. Based on previous studies revealing that children pay more heed to situational context than to emotional prosody when inferring the emotional states of others, we decided to focus on this issue, broadening the investigation to find out whether the natural combination of emotional prosody and faces (that is, paralinguistic cues) can overcome the dominance of situational context (that is, extralinguistic cues), and if so, at what age? In Experiment 1, children aged 3-9 years played a computer game in which they had to judge the emotional state of a character, based on two sources of information (that is, extralinguistic and paralinguistic) that were either congruent or conflicting. In Condition 1, situational context was compared with emotional prosody; in Condition 2, situational context was compared with emotional prosody combined with emotional faces. In a complementary study (Experiment 2) the same 3-year-olds performed recognition tasks with the three cues presented in isolation. Results highlighted the fundamental role of both cues, as a) situational context dominated prosody in all age groups, but b) the combination of emotional facial expression and prosody overcame this dominance, especially among the youngest and oldest children. We discuss our findings in the light of previous research and theories of both language and emotional development.