Xylella fastidiosa is an insect-transmitted bacterial plant pathogen found across the Americas and, more recently, worldwide. X. fastidiosa infects plants of at least 563 species belonging to 82 botanical families. While the species X. fastidiosa infects many plants, particular strains have increased plant specificity. Understanding the molecular underpinnings of plant host specificity in X. fastidiosa is vital for predicting host shifts and epidemics. While there may exist multiple genetic determinants of host range in X. fastidiosa, the drivers of the unique relationships between X. fastidiosa and its hosts should be elucidated. Our objective with this study was to predict the ancestral plant hosts of this pathogen using phylogenetic and genomic methods based on a large data set of pathogen whole-genome data from agricultural hosts. We used genomic data to construct maximum-likelihood (ML) phylogenetic trees of subsets of the core and pan-genomes. With those trees, we ran ML ancestral state reconstructions of plant host at two taxonomic scales (genus and multiorder clades). Both the core and pan-genomes were informative in terms of predicting ancestral host state, giving new insight into the history of the plant hosts of X. fastidiosa. Subsequently, gene gain and loss in the pan-genome were found to be significantly correlated with plant host through genes that had statistically significant associations with particular hosts. IMPORTANCE Xylella fastidiosa is a globally important bacterial plant pathogen with many hosts; however, the underpinnings of host specificity are not known. This paper contains important findings about the usage of phylogenetics to understand the history of host specificity in this bacterial species, as well as convergent evolution in the pan-genome. There are strong signals of historical host range that give us insights into the history of this pathogen and its various invasions. The data from this paper are relevant in making decisions for quarantine and eradication, as they show the historical trends of host switching, which can help us predict likely future host shifts. We also demonstrate that using multilocus sequence type (MLST) genes in this system, which is still a commonly used process for policymaking, does not reconstruct the same phylogenetic topology as whole-genome data.