To develop and validate an MRI-based model for predicting postoperative early (≤2 years) recurrence-free survival (RFS) in patients receiving upfront surgical resection (SR) for beyond Milan hepatocellular carcinoma (HCC) and to assess the model's performance in separate patients receiving neoadjuvant therapy for similar-stage tumors.This single-center retrospective study included consecutive patients with resectable BCLC A/B beyond Milan HCC undergoing upfront SR or neoadjuvant therapy. All images were independently evaluated by three blinded radiologists. In patients receiving upfront SR, an MRI-based Early Recurrence Outside Milan (EROM) score was developed and validated for predicting early RFS via Cox regression analyses and compared with the BCLC staging system. In separate patients undergoing neoadjuvant therapy, interval tumor progression rate and postoperative early RFS were compared between EROM-predicted high- and low-risk groups.279 patients (median, 56 years; 236 men) were included, 220 (78.9 %) undergoing upfront SR and 59 (21.1 %) received transarterial chemoembolization-based neoadjuvant therapy. Alpha-fetoprotein > 20 ng/mL (HR, 2.03; P = 0.007), size of the largest tumor (HR, 1.10; P = 0.016), infiltrative appearance (HR, 2.20; P = 0.032), and < 50 % arterial phase hyperenhancement (HR, 1.74; P = 0.023) formed the EROM score, with superior testing dataset C-index than the BCLC system (0.69 vs. 0.52, P < 0.001). The EROM-predicted high-risk (>15.3 points) patients had higher tumor progression (25.0 % vs. 0.0 %, P = 0.033) and lower postoperative 2-year RFS (16.0 % vs. 39.3 %, P = 0.025) rates after neoadjuvant therapy.In patients with resectable beyond Milan HCC, EROM allowed noninvasive prediction of postoperative early RFS and informed interval tumor progression risks after neoadjuvant therapy.Copyright © 2025 The Authors. Published by Elsevier B.V. All rights reserved.