Sigh. That's what I get for not checking my arithmetic. You're right, of course. I stand corrected on this example, but if you change the example to be O(1) at 1 year vs. O(N^2) at 1ns*N^2, the size of the dataset for the second to become slower becomes a lot larger. (Specificly, around 178 million items.) Also, there's the consideration of O() notation being the worst-case senerio. For example, even though bubble-sort is O(N^2), for nearly-sorted input, it can be quit efficent.