A number of statistical measures of the differential diffusion of multiple scalars driven by uniform mean gradients are studied by direct numerical simulation (DNS), in isotropic turbulence using up to5123grid points. Consistent with a gradient correlation given by a function of the molecular diffusivities alone, the coherency spectrum is found to be insensitive to the Reynolds number when scaled by Kolmogorov variables in wave number space. Analytical results based on joint Gaussianity assumptions are derived for important unclosed terms in the probability density function (PDF) and conditional moment closure (CMC) approaches, and shown to be successful in describing the behavior of conditional statistics extracted from DNS. The multiscalar conditional diffusion in the joint PDF equation is linear in the sample space variables and specified by the two-scalar correlation coefficients. A linear result is also derived for the term “ey,” which represents the transport of conditional fluctuations in the CMC method. The functional form obtained indicates that closure for this quantity remains an important issue at high Reynolds numbers. The transport of conditional fluctuations due to unsteady, convective and diffusive effects are analyzed separately; results for conditioning on the more diffusive versus less diffusive scalar differ in a manner traceable to the fact that less diffusive scalar fields carry more small-scale spectral content. ©1998 American Institute of Physics.