Digital 3D content creation requires the ability to exchange assets across multiple software applications. For many 3D asset types, standard formats and interchange conventions are available. For material definitions, however, inter-application exchange is still hampered by different software packages supporting different BRDF models. To make matters worse, even if nominally identical BRDF models are supported, these often differ in their implementation, due to optimisations and safeguards in individual renderers. To facilitate appearance-preserving translation between different BRDF models whose precise implementation is not known (arguably the standard case with commercial systems), we propose a robust translation scheme which leaves BRDF evaluation to the targeted rendering system, and which expresses BRDF similarity in image space. As we will show, even naive applications of a nonlinear fit which uses such an image space residual metric work well in some cases; however, it does suffer from instabilities for certain material parameters. We propose strategies to mitigate these instabilities and perform reliable parameter remappings between differing BRDF definitions. We report on experiences with this remapping scheme, both with respect to robustness and visual differences of the fits.