Aim of this study was the assessment of left atrial appendage (LAA) dimensions comparing 2D- to 3D-TEE measurements in patients with nonvalvular atrial fibrillation undergoing percutaneous LAA occlusion. Patients underwent transesophageal echocardiography (TEE) before, during and 45 days after intervention. The maximal LAA orifice diameters in 2D-TEE (LODmax 2D) were obtained from multiple views. Test–retest reliability (screening vs. implantation), inter- and intra-observer variability for echocardiographic parameters were assessed by two independent examiners. Overall, 74 patients underwent percutaneous LAA occlusion. 2D-TEE significantly underestimated the maximal LAA orifice diameter compared with 3D-TEE (screening LODmax 2D 21.11 ± 2.75 mm vs. 22.52 ± 3.45 mm for LODmax 3D, p < 0.001; during implantation LODmax 2D 21.56 ± 3.48 mm vs. 22.99 ± 3.24 mm for LODmax 3D, p < 0.001). The intraobserver and interobserver variability calculated as coefficient of variation (CV) were both lower for the 3D-TEE quantification of the maximal orifice diameter (intraobserver CV for 3D-TEE 6.07 % vs. 9.31 % for 2D-TEE; interobserver CV for 3D-TEE 6.73 % vs. 9.69 % for 2D-TEE). Compared to 3D-TEE the test–retest reliability of 2D-TEE showed a lower intraclass correlation coefficient calculated as average of raters (0.92 for 3D-TEE vs. for 2D-TEE 0.78). Firstly, 2D-TEE significantly underestimates the maximal LAA orifice diameter compared to 3D-TEE. Secondly, 3D-TEE measurements are associated with a lower observer variability and higher reliability than 2D-TEE.