I think that no answer is possible with essentially no information on what you're going to do... in particular if we take into account that with SEM you can do several (very) different types of analyses.

You are able to use it considering observed variables (use rectangles in AMOS). If your variable is a categorical one, you only are able to use it if it is a dummy one. If it isn’t you can create several dummys (e.g., yes/no; 1, 0). You can run path analyses, multivariate regression analysis using observable variables.

Its useful if you find your dataset variables(/latent or constructs) have causation and you want to test or explore several paths/inter-relationship.