For nonparametric density estimation, we apply the idea of the local linear smoother of Fan (1993) to construct a new kernel density estimator. The asymptotic variance and the asymptotic bias of this density estimator are given. By these, this density estimator does not suffer from boundary effects in the case that boundary points of the support of the underlying density are known in advance. Hence it does not require modifications on boundary regions. For constructing this density estimator in practice, the choices of the kernel function and the bandwidth are considered. Application to real data demonstrates the usefulness of this density estimator.