The physical conditions that govern the applicability of noncompartmental analysis methods apparently are not all completely appreciated in the user community. This increasingly popular approach to kinetic analysis is not "model independent," as it is often called, and if it is used, for example, for substances synthesized in pools not accessible to direct test-input probes, or substances metabolized in pools inaccessible to direct measurement, results will be more or less in error. STrictly speaking, all endogenous sources (e.g., secretions) and all sinks (e.g., catabolism) each must be uniquely associated with the same (central) pool(s) into which test inputs are introduced, the same pool(s) where all measurements must be made. The consequences of not meeting these conditions are explored by evaluating the errors committed in estimating distribution volumes VD, plasma (metabolic) clearance rates PCR, and mean residence or transit times t, in mammillary-connected systems of arbitrary size (as an example), in which sources and sinks exist in one or more noncentral pools. It is shown that noncompartmental analysis always underestimates VD and t under any such conditions, and PCR is underestimated if there are noncentral sources. Formulas for the errors are given, which could be large or small depending on several complex factors not easy to discern without additional information about the system.