Thanks for your comments, unfortunately I have already spent a lot of time testing this. As you know, the performance of a hash table depends on a combination of factors, including hash table size, number of items added, hash algorithm, whether the table sizes needs to be prime/odd/even, item lengths, item content, and so on. You cannot test these in isolation. In your tests, FNV1A performed better/worse at certain table sizes. That is simply how it worked out for the type and amount of data you used. Remember, no other hash table code was changed, just the hash algorithm. I may in future add a switch to the /hmake command that allows you to set the hash algorithm. However, for now, the beta has reverted to the older hash algorithm.