Errors in three dimensions

Now that some protein X-ray structures have been proved to contain major errors, the question of the precision of 3-dimensional structures is taken seriously by crystallographers and NMR spectroscopists. Errors which cannot be avoided during model building in electron density maps, should correct themselves during crystallographic refinement, and the precision of the refined model should reach 0.15 to 0.25 A depending on the resolution of the data. Independent estimates based on homologous protein structures confirm that better than 0.5 A precision is commonly achieved, at least for C alpha and main chain atoms. The precision of NMR structures is less easily evaluated, but it should be better than 2 A when a sufficient number of NOE distance constraints are available. One may deplore the fact that not all published structures meet these standards, but possible errors should not be an excuse for not depositing atomic co-ordinates in data banks.