The dipole moment is a simple descriptor of the charge distribution and polarity and is important for understanding and predicting various molecular properties. Semiempirical (SE) methods offer a cost-effective way to calculate dipole moment that can be used in high-throughput screening applications although the accuracy of the methods is still in question. Therefore, we have evaluated AM1, GFN0-xTB, GFN1-xTB, GFN2-xTB, PM3, PM6, PM7, B97-3c, HF-3c, and PBEh-3c SE methods, which cover a variety of SE approximations, to directly assess the performance of SE methods in predicting molecular dipole moments and their directions using 7211 organic molecules contained in the QM7b database. We find that B97-3c and PBEh-3c perform best against coupled-cluster reference values yielding dipole moments with a mean absolute error (MAE) of 0.10 and 0.11 D, respectively, which is similar to the MAE of density functional theory (DFT) methods (∼0.1 D) reported earlier. Analysis of the atomic composition shows that all SE methods show good performance for hydrocarbons for which the spread of error was within 1 D of the reference data, while the worst performances are for sulfurcontaining compounds for which only B97-3c and PBEh-3c show acceptable performance. We also evaluate the effect of SE optimized geometry, instead of the benchmark DFT geometry, that shows a dramatic drop in the performance of AM1 and PM3 for which the range of error tripled. Based on our overall findings, we highlight that there is an optimal compromise between accuracy and computational cost using GFN2-xTB (MAE: 0.25 D) that is 3 orders of magnitude faster than B97-3c and PBEh-3c. Thus, we recommend using GFN2-xTB for cost-efficient calculation of the dipole moment of organic molecules containing C, H, O, and N atoms, whereas, for sulfur-containing organic molecules, we suggest PBEh-3c.