Asbestos, thalidomide, and smog in the environment have all given spectacular evidence of the power of man-made chemical substances to harm people. Phenobarbitone, paracetamol, DDT and penicillin are chemicals that have given large benefits for small risk. Epidemiological evidence allows us to consider dose, response and cost for some of these. For new chemicals we try to assess risk before human exposure starts, using model test systems. The tests are based on studies of substances known to have caused harm to man. The methods of risk and benefit assessment are not perfect, so that post-exposure surveillance is necessary for new and old drugs, pesticides and industrial and environmental chemicals. Risk assessment then takes the form of a continuous review of chemical use, as the relative risks, costs, benefits and alternatives change with developing technology. Those at the exposure end may have interests that conflict with the interests of the rest of society, so that any evaluation process for the risks should include a process for winning the agreement of the people at risk. The case of 2,4,5-T shows how the scientifically difficult and socially fragile process of risk assessment and evaluation can easily be disrupted. We need more systematic processes of quantification and more robust processes for evaluation of risks from chemicals.