The study developed a deep learning model based on preoperative CT images to predict local recurrence-free survival in patients with primary retroperitoneal sarcoma. They enrolled 115 patients from 2013–2024, divided into a training set (86 patients) and a validation set (29 patients). The DL-score model outperformed the radiomic model (Rad-score) and the clinical model with a C-index of 0.778 in training (vs. 0.716 and 0.721) and 0.483 in validation. DL-score was an independent predictor (adjusted HR = 5.950, 95% CI: 2.800–12.644; p < 0.001) and divided patients into high- and low-risk groups (p < 0.0001). The combined DLCM model achieved a C-index of 0.848 (95% CI: 0.790–0.915) in training and 0.749 (95% CI: 0.601–0.878) in validation. The model showed good calibration and clinical utility for risk stratification. The DL model enables preoperative prediction and supports individualized treatment.