Experimental observations of the dissolution of calcium sulfate by flowing water have been used to investigate the assumptions underlying pore-scale models of reactive transport. Microfluidic experiments were designed to observe changes in size and shape as cylindrical disks (radius 10 mm) of gypsum dissolved for periods of up to 40 days. The dissolution flux over the whole surface of the sample can be determined by observing the motion of the interface. However, in order to extract surface reaction rates, numerical simulations are required to account for diffusional hindrance across the concentration boundary layer; the geometry is too complex for analytic solutions. We have found that a first-principles simulation of pore-scale flow and transport, with a single value of the surface reaction rate, was able to reproduce the time sequence of sample shapes without any fitting parameters. The value of the rate constant is close to recent experimental measurements but much smaller than some earlier values. The shape evolution is a more stringent test of the validity of the method than average measurements such as effluent concentration, because it requires the correct flux at each point on the sample surface.