Latitudinal and field-aligned cosmic ray gradients have been measured separately by inter-normalization of the similar integral response (greater than approximately 30 MeV/nucleon) and high counting rates of approximately 60 counts per second of the anticoincidence detectors of the LECP experiment on Voyagers 1 and 2 and the CPME experiment on IMP 8. The unambiguous separation of latitudinal from field-aligned gradients in long-lived (greater than 10 days) cosmic ray structures is possible because during their Earth-Jupiter transits, the two Voyagers are never separated by more than a few degrees in latitude or longitude, or by more than 8% of the helioradius of either spacecraft, and because the Voyagers remain magnetically well-connected to Earth, which allows a direct estimate of field-aligned gradients. The IMF connection is estimated by using the solar wind velocities measured by the PLS experiment on Voyager 1. Latitudinal gradients of approximately 2 to 5%/deg are found in short-lived (10 to 30 days) structures, while they are approximately 1 to 2%/deg in structures recurring over several solar rotations. Radial gradients, except in the onsets of Forbush decreases, are commonly approximately 2%/AU, although there are rotations on which neither a radial nor a latitudinal gradient is measurable above 1%/AU or 1%/deg. Application of diffusion-convection theory to these gradients shows that if one assumes that diffusion dominates transport transverse to the magnetic field, one obtains an upper bound on the transverse mean free path for scattering &lgr;⊥ much less than 10-3 AU at 1 GeV/nucleon, which is inconsistent with values predicted by diffusion theory. The contrary hypothesis of very weak scattering is theoretically self-consistent and is also supported by the observation of the effects expected from small-scale (a few degrees) latitudinal gradients upon the diurnal variation of high latitude neutron monitors.