Résumé : Objectives: To describe the pitfalls encountered in switching over from conventional smears (CSm) to liquid-based cytology (LBC). To explore modifications of our usual diagnostic criteria. Study Design: 190 ThinPrep breast samples with paired biopsies were retrospectively evaluated by two breast cytopathologists experienced only in CSm. They were again studied after LBC training. The diagnostic performances were compared. In additional calculations we included those C4suspicious samples containing 70% high-grade nuclei and up to 30% nonmalignant cells in the C5positive category, simulating that they harbored a malignant one-cell population. We prospectively validated this modification of our diagnostic criteria. Results: Training resulted in higher complete sensitivity: 94 versus 86% (p value 0.003) and lower false negative ratio: 4 versus 12% (p value 0.003). Training generated higher complete sensitivity than collaboration without training: 94 versus 89% (p value 0.008). In the simulation, the modified criteria increased absolute sensitivity to 74% with a 0.6% false positive rate. In the validation series, they generated up to 91% absolute sensitivity, 12% suspicious rate and no false negative and false positive diagnoses. Conclusion: Training in breast LBC may increase diagnostic performance. Samples containing 70% high-grade nuclei or more can be categorized as malignant. Copyright © 2013 S. Karger AG, Basel.