Abstract Aim Alpha‐fetoprotein (AFP) checkup with abdominal ultrasonography for hepatocellular carcinoma (HCC) surveillance remains controversial. We evaluated a serial AFP‐increase and high AFP levels in the prediction of HCC. Methods At‐risk patients with chronic liver disease underwent HCC surveillance with trimonthly AFP measurement were included and categorized into HCC and non‐HCC groups. Their AFP levels at 12, 9, and 6 months (−6M) before the outcome date were evaluated. Group‐based trajectory analysis and multivariable regression analysis were performed to identify AFP trajectories as risk predictors for HCC. Results Overall, 2776 patients were included in the HCC ( n  = 326) and non‐HCC ( n  = 2450) groups. Serial AFP levels were significantly higher in the HCC than the non‐HCC groups. Trajectory analysis identified AFP‐increase group (11%) increased 24‐fold risks of HCC compared with the AFP‐stable (89%) group. Compared with patients without the AFP‐increase, a serial 3‐month AFP‐increase ≥10% elevated HCC risk by 12.1‐fold (95% CI: 6.5–22.4) in 6 months, and the HCC risks increased 13–60 fold in patients with cirrhosis, hepatitis B, or C receiving antiviral therapy, or AFP levels <20 ng/ml. Combining serial AFP‐increase ≥10% and AFP ≥20 ng/ml at −6M significantly increased 41.7‐fold (95% CI: 13.8–126.2) HCC risks. In patients who underwent biannual AFP checkups, those with both 6‐month AFP‐increase ≥10% and AFP ≥20 ng/ml increased 22.1‐fold (95% CI: 12.52–39.16) HCC risks in 6 months. Most HCCs were detected at an early stage. Conclusions Serial 3–6‐month AFP‐increase of ≥10% previously and AFP level of ≥20 ng/ml significantly increased HCC risks in 6 months.