In this article, single and multiple regression functions based on osteological measurements were obtained from a large data set of extant flying birds for estimating the body mass of 42 Mesozoic specimens from stem taxa Archaeopterygidae, Jeholornithidae, Sapeornithidae, Confuciusornithidae, and Enantiornithes, and basal members of Ornithuromorpha. Traditionally, body mass has been estimated in fossil vertebrates using univariate scaling functions. In contrast, multiple regression functions have been used less frequently. Both predictive methods can be affected by different sources of error from statistics, phylogenetic relationships, ecological adaptations, and bone preservation; however, although some studies have addressed these biases, few have tested them within the context of a single data set. In our data set, we find that the models with greater predictive strength and applicability for new specimens, especially for stem taxa, are those derived from multiple regression analyses. For this reason, we suggest that multiple regression analyses may provide improved predictive strength for stem group specimens. Moreover, the methodology used for selecting variables allowed us to obtain specific sets of predictors for each fossil stem group that presumably minimized the variation resulting from historical contingency (i.e. differences in skeletal morphology arising from phylogeny), locomotor adaptations, and diagenetic compaction. The loss of generalizability in the multiple regression models resulting from collinearity effects was negligible on the body mass estimates derived from our data set. Therefore, the body mass values obtained for Mesozoic specimens are accurate and can be used in future studies in a number of palaeobiological and evolutionary aspects of extinct birds, particularly the first stages of avian flight.