question on pca based face recognition

I have a questioner regarding PCA/Eigenface classifier,

1. Shall I calculate the eigen-vectors first in case of input images, then train the face matrix & then calculate the dist between test and train images for final classification?

                            OR

2.Shall I train the face matrix first & then calculate the eigenvectors then calculate the dist between test and train images for classifier?

Please reply & thanks in advance

[Walter Roberson]

Does this question replace your older question http://www.mathworks.co.uk/matlabcentral/answers/56710-reconstruction-of-an-image-by-pca ?