To investigate whether somatising tendency, low mood and poor self-rated health (SRH) predict incident arm pain, and whether these factors and beliefs about causation and prognosis predict symptom persistence, we conducted an 18-month postal follow-up in 1798 working-aged subjects, sampled from the registers of five British general practices. At baseline questions were asked about pain in the arm (lasting >or=1day in the prior 12months), mental health (Short-Form 36 (SF-36MH)), somatising tendency (the Brief Symptom Inventory (BSI)), SRH, and beliefs about causation and prognosis. At follow-up we asked about arm pain in the last four weeks, and whether it had been present on >or=14days. Associations with incidence and persistence were explored using logistic regression. The 1256 participants (70% response) comprised 613 free of, and 643 with, arm pain initially. Among the former, 21% reported new pain at follow-up, while 53% of the latter reported symptom persistence. The odds of both incident and persistent arm pain were significantly raised (1.7- to 4-fold) in the least vs. most favourable bands of SF-36MH, BSI and SRH. Even stronger associations were found for arm pain on >or=14days. Persistent pain was significantly more common among those who attributed their pain to work or stress, and in those who expected symptoms still to be a problem in 12months. Thus, SRH and mental health indices were strong predictors of incident and persistent arm pain in adults from the community, while persistence was also predicted by beliefs about causation and prognosis.