Rules suffer from two serious defects. The world is complex, and so the creation and application of rules is difficult; and it changes, so that rules become obsolete. In recent years, the conventional wisdom on financial regulation has shifted away from reliance on rules and back towards a ‘supervisory approach’, in which regulators rely more on banks’ own estimates of risk, and focus more on banks’ risk management systems and controls than on their compliance with crude rules. The Basel Committee’s proposals for a new Capital Accord (‘Basel 2’) follow this approach. In this paper I identify four problems with this approach. First, relying on banks’ estimates is not a solution to the problems caused by externalities. Secondly, for supervision to be effective, supervisors must have the skills, incentives and legal powers to change banks’ behaviour. It is difficult and costly to design a regime in which supervisors have desirable incentives. The supervisory approach appears ill-suited to the circumstances of developing countries, at least. Thirdly, the supervisory approach is based on qualitative standards and general principles. This delegates a great deal of discretion to bureaucrats, which is legally and politically difficult in many countries. Fourthly, the implementation of standards is essentially unobservable. As a result, the international regime will shift significantly towards decentralisation. An alternative approach would be to retain an emphasis on quantitative rules, and to improve the process for interpreting, enforcing and revising them.