The post-earthquake damage probability of urban building clusters is strongly correlated with local site conditions. Hence, this study employs a semiempirical method to investigate this relationship for the rapid and accurate prediction of damage probability for urban small- and medium-scale building clusters after strong earthquakes. A specialized database was established by compiling field-observed building damage data and corresponding site vulnerability index K-value from destructive earthquakes. With the use of 70% randomly selected data from the database, a probability model for slight, moderate, and severe damage in building clusters after strong earthquakes was developed to predict the multilevel damage probabilities of urban small- and medium-scale building clusters. The remaining 30% data served as a test set to validate the damage probability model, with the results demonstrating high predictive accuracy. Application of the model to 16 building clusters in the Saihan District, Hohhot, further confirmed its feasibility for regional damage probability analysis.