Rapid on-site evaluation (ROSE) is a procedure that allows immediate assessment of adequacy of cytological specimens obtained by fine needle aspiration (FNA). The application of ROSE diagnostic categories has been applied in various organs, but not in thoracic pathology. We aimed to retrospectively assess the concordance with the final diagnosis of a categorization from C1 (inadequate) to C5 (neoplastic) during ROSE performed with bronchoscopic or percutaneous sampling procedures of thoracic lesions in a large series of consecutive cases. This retrospective single-center study evaluated 2282 consecutive ROSEs performed on 1827 patients from January 2016 to December 2020 in 994 cases of transbronchial needle aspiration (TBNA) in peripheral pulmonary lesions, in 898 transthoracic FNAs, in 318 ultrasound-guided TBNAs, in 50 conventional TBNAs and in 22 endobronchial TBNAs. False positive and false negative cases of ROSE were 43 (1.88%) and 73 (3.2%), respectively, when compared with the definitive diagnosis. The sensitivity, specificity and the positive and negative prognostic values of ROSE were 94.84%, 95.05%, 96.89% and 91.87%, respectively. Overall concordance between ROSE and the final diagnosis was 0.8960 (Cohen’s kappa). No significant differences were observed in terms of sampling procedures and type and location of the lesions. A tiered classification scheme of ROSE from C1 to C5 during bronchoscopic and percutaneous sampling procedures is helpful in effectively guiding clinical management of patients with thoracic lesions.