Outliers are observations which are very unlikely to be correct, as judged
by independent observations or other prior information. Such unexpected obs
ervations are treated, effectively, as being more informative about possibl
e models, so they can seriously impede the course of structure determinatio
n and refinement. The best way to detect and eliminate outliers is to colle
ct highly redundant data, but it is not always possible to make multiple me
asurements of every reflection. For Iron-redundant data, the prior expectat
ion given either by a Wilson distribution of intensities or model-based str
ucture-factor probability distributions can be used to detect outliers. Thi
s captures mostly the excessively strong reflections, which dominate the fe
atures of electron-density maps or, even more so, Patterson maps. The outli
er rejection tests have been implemented in a program, Outliar.