OBJECTIVE The aim of the study was to compare the incidence of anaemia in patients treated with zidovudine (ZDV) with that in patients treated with highly active antiretroviral therapy (HAART) not including ZDV. METHODS Using HIV Insight, a database of abstracted US HIV care centre medical charts, ZDV-naïve patients starting ZDV-containing HAART were compared with those starting non-ZDV, nucleoside reverse transcriptase inhibitor-containing HAART. Cohorts were divided as follows: group 1: without baseline anaemia [haemoglobin (Hb) >or=11 g/dL]; group 2: with baseline anaemia (Hb <11 g/dL). The incidence of anaemia (anaemia diagnosis, Hb <11 g/dL, erythropoietic therapy or blood transfusion) was computed for group 1. The anaemia hazard ratio (HR) was adjusted using Cox regression. The rate of worsening anaemia (Hb decrease >or=1.0 g/dL) was computed for group 2. RESULTS In group 1, the incidence of anaemia was 24.3 and 8.1 per 100 person-years in the ZDV and non-ZDV cohorts, respectively, after 6 months of follow-up, and 12.5 and 5.3 per 100 person-years after 24 months. Significant predictors of anaemia were ZDV, low initial Hb, injecting drug use, CD4 count <200 cells/microL and AIDS. The adjusted HR for ZDV was 1.6 (P=0.005). In group 2, the ZDV/non-ZDV risk ratio for worsening anaemia was 2.2 (95% confidence interval 1.1-4.3). CONCLUSIONS Patients initiating ZDV-containing HAART are at greater risk of developing new anaemia or worsening anaemia than patients initiating non-ZDV-containing HAART.