An accurate prognostic model for acute myeloid leukemia (AML) can guide personalized treatment. In our prospective cohort of 591 patients newly diagnosed with AML, we evaluated the prognostic significance of serum albumin levels. We recognized baseline serum albumin as a prognostic factor by univariate Cox regression analysis (albumin-high vs albumin-low: overall survival [OS]: hazard ratio [HR]: 0.679, 95% confidence interval [CI]: 0.529–0.870, P = .002; cumulative incidence of relapse [CIR]: HR: 0.705, 95% CI: 0.530–0.938, P = .017) and multivariate Cox regression analysis (OS: HR per g/L: 0.966, 95% CI: 0.940–0.993, P = .014; CIR: HR per g/L: 0.959, 95% CI: 0.927–0.993, P = .017). In the subgroup analysis, serum albumin was prognostic significant in patients who received intermediate-dose cytarabine combined with daunorubicin and omacetaxine mepesuccinate induction (albumin-high vs albumin-low: OS: HR: 0.585, 95% CI: 0.397–0.863, P = .007; CIR: HR: 0.551, 95% CI: 0.353–0.861, P = .009) rather than those receiving conventional-dose induction regimens. In addition, the impact of baseline serum albumin level was evident in patients with intermediate European LeukemiaNet risk (albumin-high vs albumin-low: OS: HR: 0.617, 95% CI: 0.424–0.896, P = .011; CIR: HR: 0.617, 95% CI: 0.388–0.979, P = .040). Gene set enrichment analysis revealed that leukemia stem cell signatures were enriched in patients with low serum albumin levels. Our study suggested that baseline serum albumin level was associated with the inherent properties of AML and correlated with patient outcomes.