In Denes and Keedwell's book the word "latin" is not capitalized, and there seems to be some precedent in the literature for this usage. However, the vast majority of work on the subject capitalizes the term "Latin square."
Indeed, most English dictionaries and computer spell checkers treat the word Latin as a proper noun. To me, this doesn't seem a particularly compelling argument for its usage in mathematics as, for example, neither Merriam Webster nor Dictionary.com contain an entry for the word "quasigroup."
Perhaps this is a pedantic question, but I have encountered it in personal correspondence several times in the past couple of months. Nonetheless, it may also be worth asking if there is any value to having a consistent convention within the mathematical community.