Early diagnosis of hepatocellular carcinoma (HCC) is the key to the delivery of effective therapies. The conventional serological diagnostic test, estimation of serum alpha-fetoprotein (AFP) lacks both sensitivity and specificity as a screening tool and improved tests are needed to complement ultrasound scanning, the major modality for surveillance of groups at high risk of HCC. We have analysed the serum proteome of 182 patients with hepatitis C-induced liver cirrhosis (77 with HCC) by surface-enhanced laser desorption/ionisation time-of-flight mass spectrometry (SELDI). The patients were split into a training set (84 non-HCC, 60 HCC) and a 'blind' test set (21 non-HCC, 17 HCC). Neural networks developed on the training set were able to classify the blind test set with 94% sensitivity (95% CI 73-99%) and 86% specificity (95% CI 65-95%). Two of the SELDI peaks (23/23.5 kDa) were elevated by an average of 50% in the serum of HCC patients (P<0.001) and were identified as kappa and lambda immunoglobulin light chains. This approach may permit identification of several individual proteins, which, in combination, may offer a novel way to diagnose HCC.