Data currently available in primary care Electronic Patient Records (EPR) can potentially be used to study quality of care. In this paper we investigate to which extend these data can reflect GPs' "thoughts" that are an important issue when considering GPs' practice and quality improvement cycle. Within the Resoprim project, we mainly used the consolidated data of three software systems, 26 practices, 1,554 hypertensive patients and 1,977 contacts. Extracted data from the EPR were: some diagnoses, some drugs, referral events, marital status, some parameters (smoking status, height, weight, blood pressure). As "gold standard" of GPs' thoughts we used an electronic questionnaire at the end of each contact. Measures of missing and incoherent values were used to assess our "gold standard". Sensitivity, positive predictive values, correctness and global completeness were used to measure the quality of the automatic extracted data (our proxy). For the "gold standard", the global percentage of missing values is 1.88% and of incoherent values is 3.92%. For most of the practices, the PPV or the correctness of automatic extracted drugs and automatic extracted parameters is high (>95%). The PPV of automatic extracted diagnoses is variable (42.1% to 94.9%). The sensitivity of automatic extracted diagnoses and drugs is lower than 67%. For most of the practices the sensitivity of automatic extracted parameters (excl. smoking status) is higher than 95%. The global completeness of height and weight is lower than 76%. Referrals are badly recorded or extracted. Currently in Belgium, without additional investigations, databases built on data extracted from EPRs can hardly be considered as good proxies of what is thought or known by the GPs. To use them as proxies, we should at least develop tools such as electronic questionnaires to calibrate them. As priority, we suggest an improvement of the extraction procedure design, of the current software interfaces and of the quality control of the extraction modules in order to improve respectively the extracted drugs sensitivity, the global completeness of extracted parameters and the PPV of extracted diagnoses. Training GPs could also be helpful.