1: \begin{abstract}

2: We suggest approximating the distribution of the sum of independent and

3: identically distributed random variables with a Pareto-like tail by

4: combining extreme value approximations for the largest summands with a

5: normal approximation for the sum of the smaller summands. If the tail is

6: well approximated by a Pareto density, then this new approximation has

7: substantially smaller error rates compared to the usual normal approximation

8: for underlying distributions with finite variance and less than three

9: moments. It can also provide an accurate approximation for some infinite

10: variance distributions.

11: 

12: \textbf{AMS 2000 subject classification: }60F05, 60G70, 60E07

13: 

14: \textbf{Key words and phrases: }Regular variation, rates of convergence

15: \end{abstract}

16: