It is theoretically plausible that social anxiety (SA) and social relationships (SR) can influence each other. However, the available empirical evidence is inconsistent, leading to substantial uncertainty regarding the cross-lagged relations between SA and SR. This meta-analysis systematically integrates data from 107 longitudinal studies, comprising 110 independent samples and involving a total of 115,133 participants from childhood to adulthood. Four types of SR were assessed: family-related, school-related, romantic, and general relationships. One-stage meta-analytic structural equation modeling was applied to fit four cross-lagged panel models and to test potential moderators. No significant publication bias was detected. Effect size analyses revealed that prior SA significantly and negatively predicted quality of all types of SR. Family-related and general relationships each predicted prospective SA symptoms, but school-related and romantic relationships did not. No moderators were identified in analyses of family-related and romantic relationships. However, the publication year, sample age, gender, reporter, and time lag played a moderating role in analyses of school-related and general relationships. These findings suggest that SA is a crucial factor undermining SR and that dysfunctional family and general relationships also contribute to the exacerbation of SA symptoms. The strengths, limitations, and future directions of this study are discussed.