To compliment what others have highlighted already, the file system is itself a database of sorts and often not cluttering it improves speed a slight bit and design inset a lot, especially when rather using a tool (such as a DB engine) whose very purpose it is to make data access fast and convenient. I would most certainly go straight for the single database with well-defined Indexes (there are good and bad ways to index, but that might be a different question thread). There is no evidence that it would be noticeably slower for look-ups than opening individual databases. What can happen is in the case of very many simultaneous users coupled with very high "write" activity, you could see a slowdown due to write concurrency (while WAL mode handles read concurrency just fine). Moving database accesses to different files for different processes would solve that issue at a very high inset cost in design terms, if it really is an issue. Personally, I would wait for evidence of undue slowdown before considering the multi-file option. > *(Edit by drh: fix the spelling of "SQLite" in the title.)* Oh burn! :)