In a series of studies, we have identified TFP5, a truncated fragment of p35, the Cdk5 kinase regulatory protein, which inhibits Cdk5/p35 and the hyperactive Cdk5/p25 activities in test tube experiments. In cortical neurons, however, and in vivo in Alzheimer's disease (AD) model mice, the peptide specifically inhibits the Cdk5/p25 complex and not the endogenous Cdk5/p35. To account for the selective inhibition of Cdk5/p25 activity, we propose that the "p10" N-terminal domain of p35, absent in p25, spares Cdk5/p35 because p10 binds to macromolecules (e.g., tubulin and actin) as a membrane-bound multimeric complex that favors p35 binding to Cdk5 and catalysis. To test this hypothesis, we focused on Munc 18, a key synapse-associated neuronal protein, one of many proteins copurifying with Cdk5/p35 in membrane-bound multimeric complexes. Here we show that, in vitro, the addition of p67 protects Cdk5/p35 and has no effect on Cdk5/p25 activity in the presence of TFP5. In cortical neurons transfected with p67siRNA, we also show that TFP5 inhibits Cdk5/p35 activity, whereas in the presence of p67 the activity is protected. It does so without affecting any other kinases of the Cdk family of cyclin kinases. This difference may be of significant therapeutic value because the accumulation of the deregulated, hyperactive Cdk5/p25 complex in human brains has been implicated in pathology of AD and other neurodegenerative disorders.