Aims. The aim of this study was to explore the prognostic factors for postoperative neurological recovery and survival in patients with complete paralysis due to neoplastic epidural spinal cord compression. Patients and Methods. The medical records of 135 patients with complete paralysis due to neoplastic cord compression were retrospectively reviewed. Potential factors including the timing of surgery, muscular tone, and tumour characteristics were analyzed in relation to neurological recovery using logistical regression analysis. The association between neurological recovery and survival was analyzed using a Cox model. A nomogram was formulated to predict recovery. Results. A total of 52 patients (38.5%) achieved American Spinal Injury Association Impairment Scale (AIS) D or E recovery postoperatively. The timing of surgery (p = 0.003) was found to be significant in univariate analysis. In multivariate analysis, surgery within one week was associated with better neurological recovery than surgery within three weeks (p = 0.002), with a trend towards being associated with a better neurological recovery than surgery within one to two weeks (p = 0.597) and two to three weeks (p = 0.055). Age (p = 0.039) and muscle tone (p = 0.018) were also significant predictors. In Cox regression analysis, good neurological recovery (p = 0.004), benign tumours (p = 0.039), and primary tumours (p = 0.005) were associated with longer survival. Calibration graphs showed that the nomogram did well with an ideal model. The bootstrap-corrected C-index for neurological recovery was 0.72. Conclusion. In patients with complete paralysis due to neoplastic spinal cord compression, whose treatment is delayed for more than 48 hours from the onset of symptoms, surgery within one week is still beneficial. Surgery undertaken at this time may still offer neurological recovery and longer survival. The identification of the association between these factors and neurological recovery may help guide treatment for these patients. Cite this article: Bone Joint J 2019;101-B:872–879