CASP stands for the Critical Assessment of Techniques for Protein Structure Prediction. A number of other blogs have discussed various aspects of CASP:
Macromolecular Modeling Blog discusses CASP as well as a brief post about Fold It with a number of good links.
business|bytes|genes|molecules did a nice write up in 2006 on the fact that the PDB does not currently accept theoretical models.
In short, CASP holds a competition every other year with the goal of attaining an objective measurement of the current capabilities in protein structure prediction. The competition was started in 1994 and the results have shown improvement, however, they have begun to level off.
A number of ‘assessments’ (categories) have been examined over the years which include: high-accuracy, template based and template free modeling (more details here and at the CASP wiki).
Fold It is a game in which users compete to determine the most likely (stable) protein structure (see blog post above for details). Another approach to the protein folding problem has been the Folding@Home project that allows you to download software that bans together many computers to act as a single supercomputer (similar to Seti@Home).
The best server in automated structural prediction in 2008 was the Zhang-server (I-TASSER).
As of today, I am not particularly worried about CASP being able to put crystallographers out of a job.
In the following post, I hope to address a number of potential ways that the crystallographic community can benefit from CASP.