Etoposide-platinum (EP) chemotherapy has long been the reference treatment for grade 3 neuroendocrine neoplasms (G3 NEN). However, G3 NEN are heterogeneous, including well-differentiated tumors (NET) and poorly differentiated large (LCNEC) or small (SCNEC) cell carcinomas, whose response to EP chemotherapy varies considerably. Our aim was to evaluate predictive biomarkers for the response to EP chemotherapy in G3 NEN. We retrospectively studied 89 patients with lung (42%) and digestive (58%) G3 NEN treated by EP chemotherapy between 2006 and 2020. All cases were centrally reviewed for cytomorphology/Ki-67 and immunohistochemistry of retinoblastoma protein (Rb)/p53/p16, analyzed using a semi-quantitative score. The absence of Rb staining (Rbinap) or the absence of very intense p53 staining (p53inap) were considered inappropriate. Rb staining was also studied as a quantitative marker, the best threshold being determined by ROC curve. Intense p16 staining (p16high) also suggested cell cycle dysregulation. Our primary endpoint was the objective response rate (ORR). We included 10 G3 NET, 31 LCNEC and 48 SCNEC, which showed ORR of 20%, 32% and 75%, respectively (NET vs. NEC, p = 0.040; LCNEC vs. SCNEC, p < 0.001). The ORR was significantly higher in NEN presenting with Rbinap (63% vs. 42%, p = 0.025) and p16high (66% vs. 35%, p = 0.006). Rb < 150 optimally identified responders (AUC = 0.657, p < 0.001). The ORR was 67% in Rb < 150 (vs. 25%, p = 0.005). On multivariate analysis, only Rb < 150 was independently associated with ORR (OR 4.16, 95% CI 1.11–15.53, p = 0.034). We confirm the heterogeneity of the response to EP treatment in G3 NEN. Rb < 150 was the best predictive biomarker for the response to EP, and p53 immunostaining had no additional value.