We present a detailed study of the singlet potential energy surface for Mg(CN)(2) using a variety of ab initio computational techniques. When second-order Moller-Plesset perturbation theory is employed in conjunction with basis sets of various sizes, seven structures for Mg(CN)2 are identified as local minima: the linear isomers NCMgCN, NCMgNC, and CNMNC and the pi-complex species NCMg-pi-(CN), CNMg-pi-(CN), and Mg[-pi-(CN)](2) (two enantiomers). These isomers are connected by eight transition states to isomerization. However, while the linear structures are also found to be minima at all of the levels of theory employed here, the existence of the pi-complexes (and, consequently, of many of the transition states) is strongly level-dependent: at B3-LYP/6-31+G*, I33-LYP/6-311+G(2df), and with Hartree-Fock calculations with a variety of basis sets, none of the pi-complexes correspond to stationary points upon the potential energy surface. Furthermore, calculations employing methods designed to deliver highly accurate molecular energies (such as G2 and CBS-Q) reveal that the pi-complexes located on the MP2/6-31G* surface are higher in energy than some of the putative transition states leading to linear isomers. While a more detailed examination of partially optimized structures upon the potential energy surface (using various levels of theory including QCISD/6-311G(2df), G2, and CBS-Q, with B3-LYP/6-311+G(2df) geometries) suggests that the pi-complexes are, technically, local minima, we conclude that these pi-complexes are, at best, highly reactive intermediates on the isomerization pathways NCMgCN <-> NCMgNC and NCMgNC <-> CNMgNC and that only the linear minima (NCMgCN, NCMgNC, and CNR/IgNC) correspond to meaningful and isolable chemical entities. According to both the G2 and CBS-Q techniques, the difference between the highest transition state and the global minimum (CNMgNC) is only similar to 30 kJ mol(-1).