If the 650 km discontinuity marks a compositional boundary, asd has been suggested, then the upper and lower mantle may be convecting separately. A series of laboratory experiments on two-layered convection were made in order to determine how thermal convection interacts with a stable density discontinuity. The working fluid consisted of two superposed layers of GLOBE 1132 syrup, a glucose solution with a Newtonian viscosity which depends strongly on temperature. The initial density contrast between layers ranged from 0.5% to 8%. A uniform heat flux was supplied to the base of the lower layer. By varying the heat flux, Rayleigh numbers between 4¿104 and 1¿107 were obtained. In every case, two-layered convection was observed, but in no case did a steady state result. Instead, a slow mixing between the layers occurred, driven by viscous stresses acting on the density interface. The mixing mechanism was provided by convective eddies which entrained fluid across the discontinuity in the form of thin schlieren. Mixing continued until the density contrast across the discontinuity became small enough to permit overturning. The mixing rate was determined by monitoring changes in dye concentration in each layer. It is found that the mixing rate is governed by the bulk Richardson number Ri, a measure of the ratio of between interfacial buoyancy and viscous forces. Mixing rate data from experiments covering the range 80<Ri<3300 are consistent with a power law of the form (1/Δ&rgr;)(d/dt)Δ&rgr;≂-0.05ϵ˙ Ri-1, where Δ&rgr; is the density jump across the discontinuity and ϵ˙=(q/μHT)1/2 is the scale for convective strain rate. Applying this mixing law to the mantle indicates that mass exchange between the upper and lower mantle could occur by this mechanism at a rate of 1018--1019 kg per million years. Convectively driven entrainment across the 650 km density discontinuity can provide a mechaism for interaction between the upper and lower mantle and may be an important source of mantle heterogeneity.