Clinical interpretation of changes in serial measurements of patients' HbA1c ought to be based on the knowledge of pre-analytical, analytical and intra-individual sources of variation that affect the results. The detectable change in HbA1c percentage depends on total analytical error. Since we have previously evidenced major problems in the routine use of HPLC, we compared a highly automated glycohemoglobin assay with the reference HPLC to solve the problem. The within- and between-run coefficients of variations ranged from 0.86 to 0.93%, and 2.51 to 2.12%, respectively, for the HPLC, and from 1.07 to 0.95, and 1.61 to 0.99% for the immunoturbidimetric assay. After HbA1c-assay calibration, the quality-control survey report of duplicate determinations performed on 20 consecutive days by both the HPLC and the immunologic method provided the expected mean values of control materials. The assay of 106 blood samples showed a minor yet significant bias of the immunoturbidimetric assay toward lower HbA1c. values (p 0.0001), as previously observed, although the two determination series resulted significantly correlated (r=0.96,p=0.0001). We conclude that the immunoturbidimetric assay is surely accurate, precise, and reproducible, and represents a valid alternative to the reference HPLC assay.