Maps of the fluid flow at the Earth’s core surface are important to provide insight into the workings of the geodynamo and may place useful constraints on hydromagnetic models. Fluid motion at the top of the outer core an be deduced from magnetic field data at the surface based on some physical assumption, although no unambiguous solution to this inverse problem can be derived. The time-dependent model of the main geomagnetic field and its secular variation is used to invert the radial magnetic induction equation. Because lateral variations in ore density to drive the motion are too small to be imaged seismically we use the secular variation at the surface to infer the flow near the core-mantle boundary. We focus on the physical aspects of dynamo theory and devote particular attention to the inherent problem of non-uniqueness of the derived core flow. To determine the core-surface motion the frozen-flux hypothesis is adopted which ascribes the secular variation of the observed magnetic field on the decade time-scale entirely to the effects of advection of the core material. We also examine the possible underlying reasons for substantial differences existing between maps produced by several investigators. Based on the International Geomagnetic Reference Field we infer fluid motions below the core-mantle boundary by inverting secular variation data of the surface over the 105-years epoch from 1900 to 2005. We obtain snapshot images of core flow at 5 years intervals, which are combined to give time-averaged and time-dependent parts of the motion over this interval.