1: \begin{abstract}

2: We introduce two general non-parametric methods for recovering paths of the 

3: Brownian and jump components from high-frequency observations of a L\'evy 

4: process. The first procedure relies on reordering of independently sampled 

5: normal increments and thus avoids tuning parameters. The functionality of this 

6: method is a consequence of the small time predominance of the Brownian 

7: component, the presence of exchangeable structures, and fast convergence of 

8: normal empirical quantile functions. 

9: The second procedure amounts to filtering the increments and 

10: compensating with the final value. It requires a carefully chosen threshold, 

11: in which case both methods yield the same rate of convergence. This rate 

12: depends on the small-jump activity and is given in terms of the 

13: Blumenthal-Getoor index. Finally, we discuss possible extensions, including 

14: the multidimensional case, and provide numerical illustrations. 

15: \end{abstract}