We investigated the perception of Big Five personality traits from trait-average voices when traits were based either on speakers´ self-ratings (Exp. 1, E1) or on other perceivers’ ratings of perceived personality of the original voice samples (E2). Trait-average voices were created from a voice database of 93 speakers (40 male, 53 female) using TANDEM-STRAIGHT n-way morphing. For speaker sex, trait and for two sentences, we created voice averages based on five speakers each, who scored either high or low on the target trait. We then measured perceivers´ ability to discriminate high and low trait-averages per trait. We also assessed facial trait perception (E3) using the paradigm and the full facial composite images by Kramer and Ward (2010). In trait-average voices based on self-ratings (E1), extraversion (for female speakers) and neuroticism (for male speakers) were the only traits that could be discriminated above chance levels. For trait-average voices which were based on other perceivers´ personality ratings of individual voices (E2), all Big Five traits were discriminated with high accuracy, demonstrating stereotyping in the sense of consistent (though not necessarily valid) personality impressions from voices. By comparison with E1, we found substantially better perception of self-rated traits from faces (E3), for all traits except for openness, replicating Kramer and Ward (2010). Individual differences in trait perception were substantial, and there were small but significant correlations between facial and vocal trait perception skills in both E1 and E2. Overall, the present methodological approach may offer a promising window into personality perception from voices.