An investigation of first-order perturbation (FOP) methods for computing sample worths has been performed. It is demonstrated that conventional cell homogenization methods introduce errors into FOP reactivity calculations because the intracell structure of adjoint fluxes is neglected. These errors are such that the conventional FOP reactivity expression is not accurate to first order, even in the limit of infinitesimal perturbations. A method for avoiding such approximations is introduced and applied to the calculation of small sample worths in plate-type fast reactor critical assemblies. It is demonstrated that errors introduced by real flux weighting of cross sections are such that homogenized FOP reactivity calculations overestimate fissile material worths in several critical assemblies by 5 to 10%. It is shown that these errors arise because of the heterogeneous nature of plate-type critical assemblies, and when appropriate reactivity calculations (which account for spatial heterogeneity of adjoint fluxes) are performed, a significant fraction of the long-standing central worth discrepancy is eliminated.