Background: In ovarian cancer, serum CA125 is currently the only biomarker used in routine clinical practice and has several limitations, e.g. it is not entirely specific for ovarian malignancy. Alternative serum biomarkers have been proposed for ovarian cancer, such as mesothelin or human epididymis protein 4 (HE4). Although it is known that single marker analysis is often insufficient for diagnostic purposes or relapse prediction, no comprehensive studies on the combination of these markers have been conducted, yet. Using a multi-marker prediction model, we herein investigated the clinical usefulness of mesothelin and HE4, detected in a sequence of pre- and post-operative serum samples, in complementing the standard marker CA125. Methods: Serum mesothelin, HE4 and CA125 were quantified by Lumipulse G chemiluminescent enzyme immunoassay (Fujirebio) in a total of 432 serum samples from 90 healthy controls and 192 ovarian cancer patients (thereof 81.8% with FIGO III/IV), including 151 paired pre- and post-operative samples. Results: Pre-operative serum mesothelin and HE4 were significantly elevated compared to healthy controls with a median ratio (IQR) to controls of 2.6 and 9.6, respectively (p < 0.0001). High pre-operative levels of mesothelin or HE4 alone were significantly correlated with advanced FIGO-stage (p < 0.0001) and predicted surgical outcome (p < 0.0001). We retrospectively compiled mesothelin and HE4 levels with CA125, BRCA mutational status and surgical outcome, in order to define a multi-marker setting. Both, mesothelin and HE4 levels were independent from BRCA mutational status, however they positively correlated with CA125. Detection of pre-operative HE4 > 498pM (OR = 4.1, p < 0.0001) combined with pre-operative CA125 > 516U/mL (OR = 2.2, p = 0.001) was the best combination to predict surgical outcome (AUC = 0.815). Finally, we defined a prognostic signature, including pre-operative CA125, post-operative mesothelin, HE4 and surgical outcome, which was superior to the individual prognostic capacity of all included parameters. Our model allowed the calculation of a weighted risk score, stratifying patient at “lower risk” (score 0-4) or at “higher risk” (score 5-10). Patients with higher risk had significantly shorter PFS (19.0 months (95%CI = 16.1-21.8) vs 64.1 months (95%CI = 56.8-71.3) p < 0.0001) and OS (34.0 months (95%CI = 28.6-39.3) vs 71.2 months (95%CI = 64.6-77.7) p < 0.0001). Conclusions: This is the first study, proposing an integrated prognostic risk score for ovarian cancer patients, which could easily be implemented into routine diagnostics and may guide personalized treatment decisions. Patients at “higher risk”, as identified by this model, could potentially benefit from an intensified therapy regime, whereas patients at “lower risk” might be suitable for treatment de-escalation.