Previous research has used simulated interactions to investigate emotional and linguistic speech phenomena. Here, we evaluate the use of these simulated interactions by comparing speech addressed to imaginary speech partners produced by psychology students and actresses, to an existing study of natural speech addressed to genuine interaction partners. Simulated infant- (IDS), foreigner- (FDS) and adult-directed speech (ADS) was obtained from 10 female students and 10 female actresses. These samples were acoustically analysed and rated for positive vocal affect. Our results for affect for actresses and student speakers are consistent with previous findings using natural interactions, with IDS rated higher in positive affect than ADS/FDS, and ADS rated higher than FDS. In contrast to natural speech, acoustic analyses of IDS produced by student speakers revealed a smaller vowel space than ADS/FDS, with no significant difference between those adult conditions. In contrast to natural speech (IDS > ADS/FDS), the mean F0 of IDS was significantly higher than ADS, but not than FDS. Acoustic analyses of actress speech were more similar to natural speech, with IDS vowel space significantly larger than ADS vowel space, and with the mean FDS vowel space positioned between these two conditions. IDS mean F0 of the actress speakers was significantly higher than both ADS and FDS. These results indicate that training plays an important role in eliciting natural-like speech in simulated interactions, and that participants without training are less successful in reproducing such speech. Speech obtained from simulated interactions should therefore be used with caution, and level of experience and training of the speakers should be taken into account.