Background/Objectives: There is a considerable global population of beef cattle, with numerous small-scale groups. Establishing separate reference groups for each breed in breeding practices is challenging, severely limiting the genome selection (GS) application. Combining data from multiple populations becomes particularly attractive and practical for small-scale populations, offering increased reference population size, operational ease, and data sharing. Methods: To evaluate potential for Chinese indigenous cattle, we evaluated the influence of combining multiple populations on genomic prediction reliability for 10 breeds using simulated data. Results: Within-breed evaluations consistently yielded the highest accuracies across various simulated genetic architectures. Genomic selection accuracy was lower in Group B populations referencing a Group A population (n = 400), but significantly higher in Group A populations with the addition of a small Group B (n = 200). However, accuracy remained low when using the Group A reference group (n = 400) to predict Group B. Incorporating a few Group B individuals (n = 200) into the reference group resulted in relatively high accuracy (~60% of Group A predictions). Accuracy increased with the growing number of individuals from Group B joining the reference group. Conclusions: Our results suggested that multi-breed genomic selection was feasible for Chinese indigenous cattle populations with genetic relationships. This study’s results also offer valuable insights into genome selection of multipopulations.