Publication bias remains a major challenge in meta-analysis, posing a significant threat to the validity of synthesized evidence. The fail-safe number (FSN) is commonly employed to assess the robustness of meta-analytic findings by estimating the number of unpublished studies with null or opposing results needed to nullify an observed effect, that is, to shift a statistically significant result to a nonsignificant one. However, the performance capacity of different FSN estimators may vary across meta-analytic conditions, potentially raising concerns about their reliability and interpretability. This study presents a simulation-based evaluation of the most commonly used FSN estimators: the Rosenthal FSN and the Rosenberg FSN and several modifications. We examine their performance availability by analyzing their behavior under controlled scenarios with null, weak, and moderate true effects within selective pressure. As a case study, we focus on the standardized mean difference using Cohen’s d as the effect size measure. Additionally, we explore the usability of these methods, discussing their strengths and limitations in meta-analytic applications. The validity and applicability of commonly used rules of thumb for interpreting FSN values are also assessed. Our findings provide a clearer understanding of when and how these FSN estimators yield meaningful insights, offering guidance about their appropriate use in evaluating the robustness of meta-analyses.