A few queries have come up over time on how to get Modeller to restrain the model to an idealized beta strand, so here's a quick summary.
See http://salilab.org/modeller/FAQ6v2.html#19 for information on how to specify secondary structure, including the 'STRAND' restraint type. This adds additional restraints to those derived from the homology.
See http://salilab.org/modeller/manual6v2/node100.html for a description of the types of restraints added for secondary structure. In summary, both alpha and strand restraints add phi-psi restraints to force residue mainchain conformation into class A (for helices) or B/P (for strands). See modlib/mnch.lib for the representative dihedral angles of these classes.
Note that Modeller attempts to fulfil all restraints, but violations will often occur, particularly when conflicts between user-defined and homology-derived restraints are present. This is less of a problem with helix restraints than with strands, as helix restraints enforce a number of distance restraints in addition to the phi-psi restraints, and so are in a sense 'stronger'.
There is a bug in Modeller 6v2's generation of beta restraints; it generates phi-phi restraints rather than the correct phi-psi restraints. To correct this, take the .rsr file generated by Modeller. At the end of this file should be some lines similar to the following for your beta restraints (the following should be all on one line):
R 9 1 4 25 8 6 0 327 329 330 332 327 329 330 332 1.0000 -2.3387 2.5656 0.5271 0.3072 -0.6670
Notice that the atom indices for the phi dihedral angle (columns 9 through 12, 327 329 330 332) are the same as those for what should be psi, columns 13 through 16. Correct the atom indices for psi, write out the new .rsr file, and then use the recipe in http://salilab.org/modeller/FAQ6v2.html#11 to override Modeller's restraints with your corrected ones. If in doubt about the atom indices, defining a helix restraint on the same residue range should yield the correct values (the bug is only for strand restraints).
This bug is fixed in Modeller 7, which should be released in a few weeks. Until then, apologies for the workaround!
Ben Webb, Modeller Caretaker