Relativistic fermions that are incident on a high potential barrier can pass through unimpeded, a striking phenomenon termed the 'Klein paradox' in quantum electrodynamics. Electrostatic potential barriers in graphene provide a solid-state analogue to realize this phenomenon. Here, we use scanning tunnelling microscopy to directly probe the transmission of electrons through sharp circular potential wells in graphene created by substrate engineering. We find that electrons in this geometry display quasi-bound states where the electron is trapped for a finite time before escaping via Klein tunnelling. We show that the continuum Dirac equation can be successfully used to model the energies and wavefunctions of these quasi-bound states down to atomic dimensions. We demonstrate that by tuning the geometry of the barrier it is possible to trap particular energies and angular momentum states with increased e ciency, showing that atomic-scale electrostatic potentials can be used to engineer quantum transport through graphene.A ccording to the Dirac equation, massless fermions pass unimpeded through arbitrary potential barriers at normal incidence, a phenomenon termed Klein tunnelling 1,2 . The band structure of graphene, a single layer of carbon atoms in a honeycomb lattice, provides a solid-state analogue to this effect 3-9 which is uniquely amenable to direct scanned probe measurements. Klein tunnelling arises from two fundamental properties of graphene. First, the spectrum of states is gapless 10 -regardless of the energy of the incident electron, quantum states always exist across the barrier for the electron to transition into. Second, due to the honeycomb structure of the graphene lattice, the wavefunctions of graphene are two-component spinors with amplitudes on each of the sublattices 10 . Conservation of this sublattice isospin degree of freedom in the tunnelling process is directly responsible for the high transmissibility of electrons at normal incidence 3 . At oblique incidence, however, both reflection and transmission can occur at the barrier, with reflection becoming dominant at large incidence angles 3,4 . The unique nature of Klein tunnelling and the related phenomenon of negative refraction has been proposed as a building block to steer, and even confine, electrons by the use of cleverly designed barriers 5,6,11 . Signatures of these phenomena have been seen in transport 12-14 experiments on graphene devices incorporating electrostatic barriers. However, direct imaging of the wavefunctions at potential boundaries and atomic-scale verification of the predictions of Klein tunnelling has not yet been achieved, and is the subject of this Letter.The basic ideas for Klein tunnelling discussed above can also be applied to the case of a circular quantum well in graphene. In the well-known case of two-dimensional electron gases (2DEG) in semiconductor heterostructures, such quantum wells form bound eigenstates that can be used to completely trap an electron, creating highly tunable quantum dots 15 . For circular...