Experimental evidence and mathematical models indicate that CD4+ T–cell help is required to generate memory cytotoxicT–lymphocyte precursors (CTLp) that are capable of persisting without ongoing antigenic stimulation, and that such responses are necessary to clear an infection or to control it in the long term. Here we analyse mathematical models of simian immunodefiiency virus (SIV) replication in macaques, assuming that SIV impairs specific CD4+ T–cell responses. According to the models, fast viral replication during the initial stages of primary infection can result in failure to generate sufficient long–lived memory CTLp required to control the infection in the long term. Modelling of drug therapy during the acute phase of the infection indicates that transient treatment can minimize the amount of virus–induced immune impairment, allowing a more effective initial immune sensitization. The result is the development of high levels of memory CTLp that are capable of controlling SIV replication in the long term, in the absence of continuous treament. In the model, the success of treatment depends crucially on the timing and duration of antiretroviral therapy. Data on SIV–infected macaques receiving transient drug therapy during acute infection support these theoretical predictions.The data and modelling suggest that among subjects controlling SIV replication most efficiently after treatment, there is a positive correlation between cellular immune responses and virus load in the post–acute stage of infection. Among subjects showing less–efficient virus control, the correlation is negative.We discuss our findings in relation to previously published data on HIV infection.