Non-parametric density estimation, especially with high dimensional data sets, presentsa problem requiring powerful hardware and sophisticated algorithms. Sparse gridsare an approach that offer a computationally feasible solution. However, the usualmethods, like the standard combination technique, still struggle with the sheer numberof dimensions and data points for some data sets. Recent advances in adaptivecombination techniques present new solutions for such challenges. In this thesis onesuch method - the dimension-wise spatially adaptive refinement - will be analyzed andcompared to the standard combination technique and regular kernel density estimation.To determine its effectiveness for density estimation it will be compared to thesemethods with a variety of data sets. Furthermore, a classification-based comparison ofthe dimension-wise method and the standard combination technique will be conducted.