Body mass is a fundamental biological parameter for inferring flight performance, physiological constraints, and ecological niches in early birds. However, estimating body mass in Mesozoic birds remains challenging because fossils are often affected by taphonomic compression, deformation, and incomplete skeletal preservation. Here, we developed both univariate and multivariate regression models based on a dataset of 765 extant bird specimens to identify optimal predictors and evaluate model performance. Results show that humeral midshaft width (HW) provides the strongest predictive power among univariate models; after outlier screening of the dataset (N = 745), the HW-based univariate model achieved an R2 of 0.941. Multivariate regression further improves model fit (adjusted R2 = 0.967, N = 745) and reduces both mean prediction bias (%MPE from 7.5% to 4.1%) and prediction error magnitude [mean(|%PE|) from 31.6% to 22.0%]. Applying these models to Mesozoic bird taxa, we generated refined body-mass estimates, compared clade-level patterns and model outputs, and benchmarked our results against published models. Together, these analyses support multivariate regression as a more robust framework for paleobiological mass estimation and provide a firmer basis for future studies of early avian body-size evolution.