The objective of this study was to derive and test a score that can accurately predict the presence of left atrial or left atrial appendage thrombus (LAAT) in order to identify patients with nonvalvular atrial fibrillation who can be spared transesophageal echocardiogram (TEE) and safely cardioverted. This cross-sectional observational study including 180 individuals (37.2% women) undergoing clinical, echocardiographic and laboratory evaluation (including cardiac troponin I and C reactive protein; CRP) during an atrial fibrillation episode. LAAT was sought on TEE and predictors of this transesophageal echocardiographic finding were assessed. Based on predictors of LAAT (CRP, atrial volume, troponin, episode duration and stroke or embolism) we derived the CATES score and tested its accuracy through receiver operating curve analysis. LAAT was found in 9.4%. CHADS2 and CHA2DS2-VASc had a modest performance in predicting these changes displaying a 0.620 (c-statistic) in average. Using CATES score displayed a higher area under the curve value 0.816 for LAAT. No patients with LAAT were observed in patients with CATES scores ranging from '0' to '2', which corresponded to 49.4% (n = 89) of the sample. We developed a score that presented a very good accuracy for the detection of LAAT in our sample. Further studies in other populations, such as with bigger dimensions, are needed to validate this score and confirm its capability of selecting a very low risk group of patients that can be spared transesophageal echocardiography.