In vitro thermal denaturation experiments suggest that, because of the possibility of irreversible alterations, thermodynamic stability (i.e., a positive value for the unfolding Gibbs energy) does not guarantee that a protein will remain in the native state during a given timescale. Furthermore, irreversible alterations are more likely to occur in vivo than in vitro because (a) some irreversible processes (e.g., aggregation, "undesirable" interactions with other macromolecular components, and proteolysis) are expected to be fast in the "crowded" cellular environment and (b) in many cases, the relevant timescale in vivo (probably related to the half-life for protein degradation) is expected to be longer than the timescale of the usual in vitro experiments (of the order of minutes). We propose, therefore, that many proteins (in particular, thermophilic proteins and "complex" proteins systems) are designed (by evolution) to have significant kinetic stability when confronted with the destabilizing effect of irreversible alterations. We show that, as long as these alterations occur mainly from non-native states (a Lumry-Eyring scenario), the required kinetic stability may be achieved through the design of a sufficiently high activation barrier for unfolding, which we define as the Gibbs energy barrier that separates the native state from the non-native ensemble (unfolded, partially folded, and misfolded states) in the following generalized Lumry-Eyring model: Native State <--> Non-Native Ensemble --> Irreversibly Denatured Protein. Finally, using familial amyloid polyneuropathy (FAP) as an illustrative example, we discuss the relation between stability and amyloid fibril formation in terms of the above viewpoint, which leads us to the two following tentative suggestions: (a) the hot spot defined by the FAP-associated amyloidogenic mutations of transthyretin reflects the structure of the transition state for unfolding and (b) substances that decrease the in vitro rate of transthyretin unfolding could also be inhibitors of amyloid fibril formation. Copyright 2000 Wiley-Liss, Inc.