The susceptibility of existing buildings to earthquakes is required to be assessed since building stock consists of structures that were constructed before the development of seismic standards, whether by disregarding them or by taking into account lower seismic regulations. Damage to buildings due to earthquakes not only endanger people’s lives but also causes economic losses. Because examining a large number of buildings by employing detailed building assessment methods is computationally expensive, Rapid Visual Screening (RVS) techniques are capable of assessing large building stock. Previous studies demonstrate that accuracy of the conventional RVS methods to precisely determine buildings’ damage states is limited. Therefore, it is required to develop a new RVS method. Since machine learning is extremely competent in establishing a relationship between input parameters and the target variable, this study introduces a new machine learning-based highly accurate RVS method, that can be applied in different countries, for masonry buildings using post-earthquake building screening data of 273 masonry buildings collected after the 2019 Mugello, Italy earthquake. The developed method differs from conventional RVS methods in terms of considered parameters such as spectral acceleration, the fundamental natural frequency of buildings, and the distance to the earthquake source. By comparing calculated building damage states with identified damage states through post-earthquake inspection, the developed method’s potential efficiency has been demonstrated as 88.9% accurate.