Rheumatoid arthritis (RA) has traditionally been treated using the pyramid approach, in which non-steroidal anti-inflammatory drugs (NSAIDs) are the first-line treatment and disease-modifying anti-rheumatic drugs (DMARDs) are introduced relatively late in the disease. This approach is no longer valid. Previously regarded as a benign disease, RA is now recognized as causing substantial morbidity and mortality, as do the NSAIDs used in treatment. DMARDs are more effective in controlling the pain and disability of RA than NSAIDs, and are often no more toxic. The current treatment paradigm emphasizes early, consistent use of DMARDs. A 'sawtooth' strategy of DMARD use has been proposed, in which a rising but low level of disability triggers a change in therapy. Determining the most clinically useful DMARD combinations and the optimal sequence of DMARD use requires effectiveness studies, Bayesian approaches and analyses of long-term outcomes. Such approaches will allow optimization of multiple drug therapies in RA, and should substantially improve the long-term outcome for many patients.