How did you compare scores with loading? separately or did you create a bi-plot? I strongly recommended you to create a bi-plot. With the bi-plot both scores and loadings will have the same proportions and it will be easy to interpret the results.

Which software are you using? if you have any doubt please upload the plot picture and it will be easy to help you.

All the best,

I concur with Edenio. You have to make a biplot. The results of a PCA are usually such that the scores are in standard coordinates (variance=1; mean=0) and the loadings in principal coordinates which are scaled with the square root of the eigenvalues. Biplot are scaled such that the scales and loadings are properly scaled with respect to each other.

The scores are the inner product (thanks to the loadings being unit vectors as Pieter points out) of the transposed loadings and the sample data. This means each variable in the loading is multiplied by the corresponding value in the sample, then the products are summed to calculated the score for that loading.

If a sample exhibits an identical pattern to the loading it will square all the intensities (then scaled by any multiplier that transforms the pattern).

If a sample exhibits a pattern completely orthogonal to the loading it will give 0.

If a sample is somewhere in between it will produce a score somewhere between.

The loadings are the inner product of the inverse of the scores and the sample data, so each loading is a linear combination of the sample data weighted by the score.

The larger a score the more the sample contributes to the loading, any samples that lie closer to the origin contribute less. The bigger the contribution, the more they explain the loading.