To understand the mechanisms involved in the pathogenesis of human T lymphotropic virus type I (HTLV-I)-associated myelopathy/tropical spastic paraparesis (HAM/TSP), three in vivo phenomena which have been observed in the peripheral blood of patients and differing from that in asymptomatic HTLV-I carriers must be taken into consideration: (a) the presence of increased HTLV-I viral load, (b) a higher immune responsiveness against HTLV-I antigens, and (c) biased nucleotide substitutions in the HTLV-I pX region which indicate a decreased selection pressure for viral amino acid changes. We now propose a hypothesis which focuses on the in vivo dynamics of HTLV-I infected lymphocyte migration and which incorporates these features. In addition, the hypothesis assumes the existence of a deviation in immune surveillance for HTLV-I in the central nervous system (CNS) in spite of the presence of frequent specific immune effectors. We suggest that in the active phase of HAM/TSP, accompanied with or following autoaggressive interactions between infected lymphocytes and immunocompetent cells in the CNS, there is a consequential reflux of the infected lymphocytes to the peripheral blood. The reflux of infected cells would be expected to provide peripheral blood with tissue-derived HTLV-I proviruses which have been indulged and propagated in an immune-privileged site. This process would result in and account for the observed increase in viral load and the substitution bias in HTLV-I sequences in the peripheral blood.