OBJECTIVES To measure the reliability of a data-gathering from by analysing the concordance between observers, and to check the efficacy of the corrective measures applied to improve it. DESIGN A pre-test/post-test design, without a control. SETTING Urban Health Centre. PATIENTS AND OTHER PARTICIPANTS Random selection of 50 clinical records from the list of Type 2 diabetics in the centre's computerised files. INTERVENTIONS Alterations in the data-gathering form in order to improve concordance between two independent observers, measured by means of the Kappa index for qualitative variables and absolute concordance for quantitative ones. The quality criteria defined by the Diabetes in Primary Health Care Study Group were applied. MEASUREMENTS AND MAIN RESULTS The Kappa index average improved after the intervention from 0.71 to 0.89 (p = 0.019). There were no significant changes in concordance for quantitative variables (6 out of 8 items with 100% concordance at start and finish). CONCLUSIONS It is important to verify the reliability of the measuring instruments. Our results justify this concern: two independent observers do not always obtain the same reading from the same source. Concordance improved significantly when the form was altered, so guaranteeing greater data-gathering reliability and improving the overall quality of the research.