Multiple myeloma (MM) is a heterogeneous group of disorders both genotypically and phenotypically. Response to thalidomide-based induction therapy in newly diagnosed patients varies significantly in published clinical trials. Proteomic analysis was performed on 39 newly diagnosed MM patients treated with a thalidomide-based regimen (22 responders; 17 non-responders) using immunodepletion, 2-D DIGE analysis and mass spectrometry. Zinc-a-2-glycoprotein (ZAG), vitamin D-binding protein (VDB), serum amyloid-A protein (SAA) and b-2-microglobulin (B2M) had statistically significant higher concentrations in non-responders compared to responders, while haptoglobin (Hp) had a lower concentration. ELISAs were used to validate the candidate protein biomarkers using unfractionated serum from 51 newly diagnosed MM patients (29 responders; 22 non-responders). Using logistic regression, the best possible area under the curve (AUC) was 0.96 using ZAG, VDB and SAA in combination. Leave-one-out-cross-validation (LOOCV) indicated an overall predictive accuracy of 84% with associated sensitivity and specificity values of 81.8 and 86.2%, respectively. Subsequently, 16 of 22 thalidomide-refractory patients successfully achieved complete response or very good partial response using second-line treatment suggesting that the biomarker profile is specific to thalidomide response rather than identifying patients with MM refractory to all therapies. Using a novel panel of predictive biomarkers, the feasibility of predicting response to thalidomide-based therapy in previously untreated MM has been demonstrated.