>I suppose ...
>
>They hold some kind of index on the latitude and longitude
>They then first select a subset (the rectangle form my previous message), so they don't have to calculate the distance for every record
>And finally calculate the distance (aka radius) for each record in the subset
>
>Finally, they may store the sin or cos along with the data (longitude and latitude) - so they don't have to recalculate those values every time

Sounds similar to the approach we use in 2005. But, as it is builtin the engine, it certainly is running with much more performance. It is also much more easy to use. There is less command to work with in the SQL to obtain the results.