Osteoarthritis is a serious joint disease that causes pain and functional disability for a quarter of a billion people worldwide 1 , with no disease-stratifying tools nor modifying therapy. Here, we use primary chondrocytes, synoviocytes and peripheral blood from patients with osteoarthritis to construct a molecular quantitative trait locus map of gene expression and protein abundance in disease. By integrating data across omics levels, we identify likely effector genes for osteoarthritis-associated genetic signals. We detect stark molecular differences between macroscopically intact (low-grade) and highly degenerated (high-grade) cartilage, reflecting activation of the extracellular matrix-receptor interaction pathway. Using unsupervised consensus clustering on transcriptome-wide sequencing, we identify molecularly-defined patient subgroups that correlate with clinical characteristics. Between-cluster differences are driven by inflammation, presenting the opportunity to stratify patients on the basis of their molecular profile for tailored intervention. We construct and validate a 7-gene classifier that reproducibly distinguishes between these disease subtypes. Finally, we identify potentially actionable compounds for disease modification and drug repositioning. Our findings contribute to both patient stratification and therapy development in this globally important area of unmet need.