After the cross-validated PLS analysis, you will determine the optimal number of components. Recall that the components are linear combinations of the variables (of which you have very many!), ordered in such a way that the first component will describe most of the variation in the activity, the second most of the remaining variation, etc. The cross-validated PLS analysis will be carried out for different numbers of components. As a rule of thumb, the number of components should not exceed one-third of the number of molecules. More components would lead to a model that is overtrained - it has a better fit to the training data but the predictive power is diminished.

http://www.cmbi.ru.nl/edu/bioinf4/comfa-Prac/comfa.shtml