Abstract Background Cardiac masses (CM) are an extremely heterogeneous clinical scenario, including benign and malignant neoformations. After a first echocardiographic assessment, Cardiac Computed Tomography (CCT) together with Cardiac Magnetic Resonance (CMR) and 18-Fluorodeoxyglucose Positron Emission Tomography (18- FDG-PET) represent second line and third-line imaging techniques to determine the nature of the mass. However, data regarding their diagnostic performance and a standardized imaging algorithm are lacking. Purpose To evaluate the different roles of CCT, CMR, and PET in defining the nature of CMs and to propose an evidence-based, stepwise, diagnostic approach. Materials and methods Out of 312 patients with suspected mass from January 2000 and August 2022, we enrolled 87 patients who underwent CCT, CMR and 18-FDG-PET within a month from the initial evaluation. A definitive diagnosis was achieved by histological examination or, in case of cardiac thrombi, with radiological evidence of thrombus resolution after an appropriate anticoagulant treatment. For each imaging technique, we identified a model with the strongest predictors of malignancy at multivariate analysis and evaluated their ability to discriminate between benign and malignant neoformations. A multiple model with forwarding selection was performed to identify the strongest predictors of malignancy at CCT, CMR and 18-FDG-PET. Results CCT model included 4 variables (irregular margins, mass dimension, invasiveness, and not-hypodense lesion) with an Area Under the Curve (AUC) of 0.972, 95% Confidence Interval (CI) 0.94-1.0; CMR model included 3 parameters (invasiveness, pericardial effusion and irregular margins, AUC 0.976 with 95% CI 0.95-1.0); PET model included only cardiac maximum Standardized Uptake Value (SUVmax), with an AUC 0.87 (95% CI 0.74-0.971). When implemented with SUVmax, CCT and CMR models showed only a slight improvement in their discrimination ability (AUC 0.975 and 0.986, respectively). No statistical difference was observed between CCT and CMR models regarding their discrimination ability (AUC 0.972 vs 0.976, p=0.26). However, on a multiple model with forwarding selection evaluating CCT, CMR and PET variables, only the 3 MR parameters remained significant predictors of malignancy. Conclusion After a first echocardiographic assessment, the application of the CMR model may be the most accurate second-level investigation to discriminate between benign and malignant lesions. When CMR is not available, or the patient has contraindications to CMR, the CCT model performs similarly, and 18-FDG-PET provides a negligible advantage.