Often, I will add a single article to an existing transactional replication publication. I then start the snapshot agent to copy the new article and it's data over to the subscriber. Most of the time, the agent runs fairly quickly and only generates a snapshot of 1 article (the one that I just added). However, one out of every 4 or 5 times the snapshot agent tries to take a complete snapshot of every replicated article in the entire (very large) database. This causes all sorts of problems due to the locking it creates on the tables. I can't find a rhyme or reason as to why it does this occasionally. Does anyone have any suggestions?