Pronouns can refer to discourse entities that were introduced in the same or in a previous discourse unit; for successful pronoun resolution, the antecedent must be accessible. Focusing a potential antecedent seems to boost its accessibility when the pronoun appears in a subsequent discourse unit but lowers its accessibility when the pronoun is in the same discourse unit (‘anti-focus’ effect). The current study investigates the time course of antecedent accessibility within the discourse unit using an eye-movement monitoring experiment and an offline judgement task. Participants read short German texts in which a potential antecedent appeared within the same discourse unit as the pronoun and was either (i) not in focus, (ii) in cleft focus, or (iii) focused via a particle. While there was an online reading-time advantage for pronouns when the antecedent was clefted, the judgement data showed lower acceptability ratings for clefted antecedents, i.e. an anti-focus effect. We propose that clefting provides an initial retrieval advantage for an antecedent, and that the anti-focus effect emerges during later processing stages and perhaps only when participants engage in explicit reasoning. One important implication is that it is not always possible to equate easier antecedent retrieval with greater antecedent accessibility.