The algorithm I suggested is actually reading in parallel the data and the bin ranges. Therefore, it does just 1 comparison per line of data (plus one additional comparison each time the program crosses a bin range). The complexity is O(N). It will be very difficult to do anything better for N data records (at least if the data is sorted, which was my starting hypothesis; even if the data needs to be sorted, it is still likely to be the fastest solution if a good N Log N sorting program is used).