In order to test the accuracy of glycated hemoglobin (HbA1c) in predicting mean glycemia in HIV-infected patients, we recorded consecutive HbA1c measurements from 1238 non-HIV-infected and 112 HIV-infected patients, all devoid of any hemoglobinopathy, in a retrospective, transversal study. Mean fasting glycemia from the six previous weeks (measured-Gly) and HbA1c-estimated glycemia [HbA1c-Gly (1.85x%HbA1c-4.78) mM] were compared. Mean hemoglobin, red cell volume, serum creatinine, CD4 count, and HIV viral load from the same period were collected in HIV-infected patients. Although measured-Gly was not significantly different between non-HIV-infected (6.95+/-3.23 mM) and HIV-infected patients (6.62+/-2.42 mM), HbA1c underestimated the mean fasting glycemia by 12.3% in HIV-infected as compared to non-HIV-infected patients (p=0.0001). The difference "measured-Gly-HbA1c-Gly" was correlated with the red cell volume (p<0.0001) in HIV-infected patients. We then searched for the presence of subclinical hemolysis, a cause of both macrocytosis and reduced HbA1c levels, in HIV-infected patients. To this end, we prospectively measured serum haptoglobin in 249 consecutive samples from HIV-infected subjects without any known cause of hemolysis. A very low haptoglobin level, a marker of hemolysis, was frequent and negatively correlated with the red cell volume in these patients. Treatment with nucleoside analogues was significantly associated with macrocytosis and low haptoglobin. In conclusion, HbA1c could be inappropriately low in HIV-infected patients. Its underestimation of mean fasting glycemia could be due to an antiretroviral-induced subclinical hemolysis, but further studies are needed to explore this hypothesis. Self-monitoring of blood glucose and search for latent hemolysis should be promoted in diabetic HIV-infected patients.