The phase III CELESTIAL study demonstrated improved overall survival (OS) and progression-free survival (PFS) with cabozantinib versus placebo in patients with previously treated, advanced hepatocellular carcinoma (HCC). We analyzed outcomes by baseline alpha-fetoprotein (AFP) and on-treatment AFP changes. Patients and Methods: Serum AFP was measured every 8 weeks by blinded, centralized testing. Outcomes were analyzed by baseline AFP bifurcated at 400 ng/mL and by on-treatment AFP response (≥20% decrease from baseline at Week 8). The optimal cutoff for change in AFP at Week 8 was evaluated using maximally selected rank statistics. Results: Median OS for cabozantinib versus placebo was 13.9 versus 10.3 months [HR, 0.81; 95% confidence interval (CI), 0.62-1.04] for patients with baseline AFP <400 ng/mL, and 8.5 versus 5.2 months (HR, 0.71; 95% CI, 0.54-0.94) for patients with baseline AFP ≥400 ng/mL. Week 8 AFP response rate was 50% for cabozantinib versus 13% for placebo. In the cabozantinib arm, median OS for patients with and without AFP response was 16.1 versus 9.1 months (HR, 0.61; 95% CI, 0.45-0.84). AFP response was independently associated with longer OS. The optimal cutoff for association with OS in the cabozantinib arm was ≤0% change in AFP at Week 8 [AFP control; HR 0.50 (95% CI, 0.35-0.71)]. HRs for PFS were consistent with those for OS. Conclusions: Cabozantinib improved outcomes versus placebo across a range of baseline AFP levels. On-treatment AFP response and control rates were higher with cabozantinib than placebo, and were associated with longer OS and PFS with cabozantinib.