Prior research has identified a number of dimensions along which speakers modify referring expressions. The present study aimed to determine and describe the actual relationship existing between these production characteristics and corresponding measures of listener comprehension. Spoken descriptions were elicited from eight speakers during monolog and dialog conditions of the tangram task. In a subsequent listening experiment, 163 new subjects listened to the referring expressions from the speech corpus, and were asked to select, from an array of tangram figures, the one that was being described in the referring expression. Results revealed that the production variables most commonly documented by previous researchers collectively contributed to listener comprehension to a surprisingly small, but consistently significant, degree. Furthermore, the communicative context in which referring expressions were produced did not substantially influence their overall comprehensibility; instead, communicative context, as well as the individual characteristics of the speaker who produced the referring expression, appeared to affect which variables listeners found most useful when inferring what the speaker meant.