Iterative dual-space direct methods based on phase refinement in reciprocal space and peak picking in real space are able to locate relatively large numbers of anomalous scatterers efficiently from MAD or SAD data. Truncation of the data at a particular resolution, typically in the range 3.0-3.5 A, can be critical to success. The efficiency can be improved by roughly an order of magnitude by Patterson-based seeding instead of starting from random phases or sites; Patterson superposition methods also provide useful validation. The program SHELXD implementing this approach is available as part of the SHELX package.