We have developed a new algorithm for crystal structure prediction that tak
es a conformer of an organic molecule and produces a number of candidates o
f crystal structures for this molecule, together with a score that roughly
approximates the energy of the respective crystal structure. The prediction
method takes steric aspects (dense packing) as well as chemical interactio
ns into account. The algorithm differs from existing methods in three aspec
ts. First, we analyse a single molecule rather than a collection of identic
al molecules. The analysis yields candidates for symmetry operations that a
re suitable for making the crystal. Each of the candidates is generated by
a chemical interaction between two versions of the molecule. Second, we mod
el space discretely, currently as a mesh with size 1 Angstrom. Third, the s
coring function representing energy is derived statistically from known cry
stal structures and tabulated. Our program FlexCryst computes a list of cry
stal structures ranked according to our scoring function. The new algorithm
is currently implemented for the four space groups P1, <P(1)over bar>, P2(
1), and P2(1)2(1)2(1). The three latter space groups are widespread in natu
re. The algorithm computes structural models of acceptable quality and show
s excellent time performance. (C) 1998 Elsevier Science B.V. All rights res
erved.