We construct a modified rotated- \(Q_1\) finite element, where we replace the \(P_2\) bubbles of the Rannacher–Turek rotated- \(Q_1\) element by multi-piece linear polynomials. In 2D, we use \(\{1,x,y,|\lambda _1|\}\) as the basis on a general quadrilateral, where \(\lambda _1\) is a linear polynomial which vanishes at the middle edge \(\textbf{x}_3\textbf{x}_4\) of two opposite mid-edge nodes \(\textbf{x}_1\) and \(\textbf{x}_2\) . In 3D, we use \(\{1,x,y,z, |\lambda _1|,|\lambda _3|\}\) as the basis on a general hexahedron, where \(\lambda _1\) is a linear polynomial which vanishes at the mid plane \(\textbf{x}_3\textbf{x}_4\textbf{x}_5\textbf{x}_6\) between the two opposite mid-face nodes \(\textbf{x}_1\) and \(\textbf{x}_2\) , and \(\lambda _3\) is a linear polynomial which vanishes at the mid plane \(\textbf{x}_1\textbf{x}_2\textbf{x}_5\textbf{x}_6\) between the two opposite mid-face nodes \(\textbf{x}_3\) and \(\textbf{x}_4\) . The new rotated- \(Q_1\) finite element is shown inf-sup stable and quasi-optimal in solving the Stokes equations, on general quadrilateral and hexahedral meshes. Numerical tests in 2D and 3D show the method does converge quasi-optimally on non-asymptotic parallelogram or parallelepiped meshes, while the Rannacher–Turek rotated- \(Q_1\) element fails to converge on such meshes.