Objectives <p>To develop models including MRI features for predicting recurrence-free survival (RFS) in patients with solitary small hepatocellular carcinoma (SHCC) following curative resection according to alpha-fetoprotein (AFP) expression and for achieving risk stratification.</p> Materials and methods <p>A total of 201 solitary SHCC patients diagnosed pathologically who underwent MRI were enrolled retrospectively and divided into AFP-negative (<i>n</i> = 85) and AFP-positive (<i>n</i> = 116) SHCC groups. Clinicopathological factors, imaging features, and geometric parameters associated with RFS were determined by univariate and multivariate Cox analyses. Two prognostic models predicting postresection RFS were developed by multivariate Cox proportional hazard model, and their discriminatory ability was evaluated by the concordance index (C-index) and conducted internal validation. Kaplan‒Meier analysis with RFS as the endpoint was used to assess model performance.</p> Results <p>The predictors of RFS in AFP-negative SHCC included mosaic architecture, portal-venous phase tumor-to-right erector spinae muscle signal intensity ratio, and sphericity, whereas those in AFP-positive SHCC were liver cirrhosis, tumor growth subtype, and asphericity. The C-indexes of the models in predicting RFS in AFP-negative and AFP-positive SHCC were 0.700 and 0.717, respectively. The risk probability calculated with the models stratified the two patient subgroups into two risk groups with significantly different survival outcomes (<i>p</i> &lt; 0.001).</p> Conclusion <p>Our two prognostic models may assist physicians in stratifying risk and monitoring relapses in patients with AFP-negative or AFP-positive solitary SHCCs.</p> Key Points <p><Emphasis Type="BoldItalic">Question</Emphasis> <i>This study explored the predictors of the postresection outcomes of patients with small hepatocellular carcinoma according to the level of alpha-fetoprotein expression</i>.</p> <p><Emphasis Type="BoldItalic">Findings</Emphasis> <i>Pretreatment MRI and geometric features can serve as prognostic markers for predicting postoperative recurrence-free survival of small hepatocellular carcinoma with different alpha-fetoprotein expression</i>.</p> <p><Emphasis Type="BoldItalic">Clinical relevance</Emphasis> <i>Our models guide clinicians in administering appropriate therapeutic decisions and implementing proper postoperative monitoring regimens in different alpha-fetoprotein expression small hepatocellular carcinoma patients</i>.</p> Graphical Abstract <p></p>