Previous studies on the processing of ambiguous pronominal reference have led to contradictory results: some suggested that ambiguity may hinder processing (Stewart, Holler, & Kidd, 2007), while others showed an ambiguity advantage (Grant, Sloggett, & Dillon, 2020) similar to what has been reported for structural ambiguities. This study provides a conceptual replication of Stewart et al. (2007, Experiment 1), to examine whether the discrepancy in earlier results is caused by the processing depth that participants engage in (cf. Swets, Desmet, Clifton, & Ferreira, 2008). We present the results from a word-by-word self-paced reading experiment with Dutch sentences that contained a personal pronoun in an embedded clause that was either ambiguous or disambiguated through gender features. Depth of processing of the embedded clause was manipulated through offline comprehension questions. The results showed that the difference in reading times for ambiguous versus unambiguous sentences depends on the processing depth: a significant ambiguity penalty was found under deep processing but not under shallow processing. No significant ambiguity advantage was found, regardless of processing depth. This replicates the results in Stewart et al. (2007) using a different methodology and a larger sample size for appropriate statistical power. These findings provide further evidence that ambiguous pronominal reference resolution is a flexible process, such that the way in which ambiguous sentences are processed depends on the depth of processing of the relevant information. Theoretical and methodological implications of these findings are discussed.