Both practical and conceptual problems arise in standard theories of multicomponent diffusion. They include the sometimes arbitrary and even artificial selection of one component as the solvent, the lack of explicit dependence on the properties of the solvent, and the difficulty in formulating compositionally dependent diffusion coefficients that satisfy, material-balance constraints. Two alternative approaches based on tracer diffusion coefficients for various components nr e physically reasonable and overcome these difficulties. Both approaches predict cross-diffusion, that is, diffusion of component A as a result of a gradient in component B. Either approach can he used in systems where component diffusivities differ greatly in magnitude and where the number of chemical species is large, as in polymerizations.