The lack of a biomarker for chronic pain remains an important impediment to clinical and translational pain research. The problem stems from the multiple parallel but subtle abnormalties thought to represent the chronic pain state, yielding the emerging view of chronic pain as a ‘network disorder’. This suggests analysis approaches that aim to identify distributed patterns of data (multivariate, machine learning methods) might offer the best opportunity to discover biomarkers. Here, we performed a multi-center functional brain imaging study to record state functional brain networks resting in 41 patients with chronic back pain and 33 healthy control subjects. We calculated with functional covariance matrix from 160 regions of interest, and used Sparse Multinomial Logistic Regression to classify subjects as patient or control using a leave-one-out cross validation. Diagnostic accuracy was 91.9%, with sensitivity and specificity 90.2% and 93.9% respectively. We then used graph theoretic measures to characterise the pattern of network differences between the groups, and showed that the chronic pain state was associated with disrupted network ‘assortativity’. These data provide evidence to support an accurate functional biomarker of chronic pain, and open the door to the development of translatable biomarkers using similar methodologies in animals.