The aim of this article is to investigate gender categorisation in sequences that introduce referents by their occupational titles. The study addresses the following three questions: (1) What can the systematic patterns in authentic conversation tell us about the use of gender in Icelandic? (2) What is the distribution of labour between grammatical and referential gender? (3) Does the mismatch between grammatical and referential gender cause any problems in the conversation? The data used come from 10 hours and 22 minutes of conversation recorded between 1996 and 2018 comprising everyday conversations, phone calls, radio interviews about current affairs, moderated political debates and spontaneous speeches in the parliament. In total, 55 sequences of occupational titles were identified in the data. The analysis shows that when the gender of the referent is known to the speaker, the choice of pronoun is based on the referential gender, but when the gender is unknown or the main referent is hypothetical, the speaker uses a pronoun that matches the grammatical gender of the noun. Finally, the data show that the choice of pronoun rarely triggers repairs.