In HIV disease, the mechanisms of drug resistance are only poorly understood. Incomplete suppression of HIV by antiretroviral agents is suspected to be a main reason. The objective of this in silico study is to elucidate the pharmacokinetic origins of incomplete viral suppression, exemplified for zidovudine (AZT) as a representative of the key class of nucleoside reverse transcriptase inhibitors (NRTIs). AZT, like other NRTIs, exerts its main action through its intra-cellular triphoshate (AZT-TP) by competition with natural thymidine triphosphate. We developed a physiologically based pharmacokinetic (PBPK) model describing the intra-cellular pharmacokinetics of AZT anabolites and subsequently established the pharmacokinetic-pharmacodynamic relationship. The PBPK model has been validated against clinical data of different dosing schemes. We reduced the PBPK model to derive a simple three-compartment model for AZT and AZT-TP that can readily be used in population analysis of clinical trials. A novel machanistic, and for NRTIs generic effect model has been developed that incorporates the primary effect of AZT-TP and potential secondary effect of zidovudine monophosphate. The proposed models were used to analyze the efficacy and potential toxicity of different dosing schemes for AZT. Based on the mechanism of action of NRTIs, we found that drug heterogeneities due to temporal fluctuations can create a major window of unsuppressed viral replication. For AZT, this window was most pronounced for a 600 mg/once daily dosing scheme, in which insufficient viral suppression was observed for almost half the dosing period.