It seems that the mass indexer (org.hibernate.search.MassIndexer, Hibernate Search 4.0.0.CR2) does not clean existing index files when invoked.
I observed that if I remove a @Field annotation of a property and run the indexer again, in Luke the field is still present after reopening the index.
Only if I remove the old index files before mass indexing, the field is removed from the index.
Is this normal? I don't know Luke well, but I guess it has no cache or similar.

Thanks

The Field definition will still exist, but you should not have any instance of it. Could you try searching for documents containing such a field?

The MassIndexer does not physically delete the index files, it just deletes all instances of reindexed entities. Having some leftover unused fields has no bad effect on performance (assuming they are reasonably limited).
Deleting the Index would be harder; we would have to deal with cleanup of File descriptors, concurrent Search operations would fail badly, IndexReader pools need a restart.. and finally you might not want to delete+reindex all your entities but only a specific type.

I guess that @Field deletion will happen much more frequently on developer machines than production environments.. feel free to delete your whole index directory in such case to start fresh.