This dissertation proposes a finite element procedure for evaluating the high order strain derivatives in nonlocal computational mechanics. The superconvergent second derivative recovery methods used are proven to be effective in evaluating the Laplacian of the equivalent strain based on low order (linear) elements. Current nonlocal finite element techniques with linear elements are limited to structured meshes, while the new technique can deal with unstructured meshes with various element types. Other superconvergent patch recovery (SCP) based nonlocal approaches, such as the patch projection techniques only utilize nodal based patches to evaluate the first derivatives of the strain. The SCP technique has not yet been used for recovery of higher order strain derivatives. The proposed technique is capable of evaluating the Laplacian of the equivalent strain and has the potential for even higher order derivative recovery. The same patches can be easily utilized for error estimation and adaptive meshing for nonlocal problems. We employ two super-convergent patch options: the element based patch with all neighbors or only facing neighbors. The nonlocal strain derivatives can be recovered through either a mesh nodal averaging process or directly at the patch element quadrature points after the patch least square fitting problems are solved. Numerical examples for both strain gradient damage mechanics and strain gradient plasticity problems are given. In summary, the new finite element nonlocal computational technique based on the superconvergent second derivative recovery methods is proven to be robust in evaluating the high order strain derivatives with low order element unstructured meshes.