Let X 1 , ,X n be a random sample from a distribution on the real line with an unknown density f. We discuss the performance of the kernel density estimator of the density f. The properties of kernel estimators in cases where the density f to be estimated is sufficiently smooth are well known. Instead we focus on estimation problems where f is non-smooth, i.e. f is allowed to have a finite number of jumps or kinks. Thus the robustness properties of the kernel estimator against unfulfilled smoothness assumptions are illustrated. After a review of properties of the mean integrated squared error we present a central limit theorem for the integrated squared error. This theorem extends results of Bickel, Rosenblatt and Hall. Finally, the distance between the bandwidth minimizing the integrated squared error and the bandwidth which minimizes the mean integrated squared error is discussed.