One measure of the communicative function of gestures is to test how speakers’ gestures are influenced by whether an addressee can see them or not, that is, by manipulating visibility between participants. We question traditional dependent variables (i.e., rate measures), suggesting that they may have been insufficient for capturing essential differences in the gestures speakers use in each condition. We propose that investigating the qualitative features of gestures is a more nuanced, and ultimately more informative approach. We examined how speakers distributed information between their gestures and words, testing whether this distribution was affected by the visibility of their addressee. Twenty pairs of undergraduates took part in conversations that were either face to face (n = 10) or on the telephone (n = 10). Each speaker described a drawing of an elaborate dress to the addressee. We used a semantic feature analysis to analyze descriptions of the dress’ skirt and assessed when words or gestures contributed information about five categories pertaining to features of the skirt’s unusual shape. Although speakers’ rates of gesturing and number of words did not vary significantly between conditions, speakers contributed more information and conveyed more categories in their gestures when the addressee would see them, while words carried the informational burden when addressees would not see the gestures (p’s < .001). These results suggest that gestures serve a communicative function. The semantic feature analysis is thus an example of how to explore gestures’ qualitative features within a quantitative paradigm.