Background and aims: The albumin-to-globulin ratio (AGR) has been used to predict severity and mortality in infectious diseases. The aim of this study is to evaluate the prognostic value of the AGR in COVID-19 patients. Methods: A systematic review and meta-analysis were conducted. We included observational studies assessing the association between the AGR values upon hospital admission and severity or all-cause mortality in COVID-19 patients. In the meta-analyses we used random effect models. Risk of bias was assessed using the Newcastle-Ottawa Scale (NOS). The effect measures were expressed as mean difference (MD) and their 95% confidence intervals (CI). We performed Egger's test and funnel plots to assess the publication bias. Results: The included studies had a total of 11356 patients corresponding to 31 cohort studies. Severe COVID-19 patients had lower AGR values than non-severe COVID-19 patients (mean difference (MD), −0.27; 95% IC, −0.32 to −0.22; p < 0.001; I2 = 88%). Non-survivor patients with COVID-19 had lower AGR values than survivor patients (MD, −0.29; 95% IC, −0.35 to −0.24; p < 0.001; I2 = 79%). In the sensitivity analysis, we only included studies with low risk of bias, which decreased the heterogeneity for both outcomes (severity, I2 = 20%; mortality, I2 = 5%). Conclusions: Low AGR values upon hospital admission were found in COVID-19 patients with a worse prognosis.