Pure liquid water can be supercooled to temperatures approaching - 40ºC. Indeed, it has been estimated that if all water in the Earth's hydrosphere (10\(^{18}\) metric tonnes) were cooled to - 20ºC, the time required to observe one homogeneous ice nucleation event would exceed the age of the universe. Hence, most ice nucleation on the Earth's surface or in the atmosphere must occur by heterogeneous nucleation involving a foreign substrate. Currently there is much interest in heterogeneous ice nucleation, particularly because ice nucleation in the atmosphere leads to ice cloud formation, and consequential effects on climate. Laboratory experiments can identify substances which serve as good ice nuclei (often inorganic compounds such as AgI, mineral dust such as clays, and assorted organic matter) but, due to the short time and small length scales involved in ice nucleation, experiments provide very little information about the microscopic mechanisms of ice nucleation. Molecular dynamics simulations of model systems can fill this knowledge gap, and identify the molecular or atomistic features necessary for a substrate to effectively nucleate ice. Our systematic investigations of “good” and “bad” ice nuclei reveal how the lattice structure and detailed atomistic morphology of a surface combine to determine its ice nucleating ability.