The aim of the present investigation was to evaluate and compare data from medical records with mailed questionnaires concerning risk factors for diagnostic X-ray investigation induced thyroid cancer. This study encompassed 180 patients and 360 controls, selected from the National Population Registry. Information on medical X-ray exposure was obtained by medical records and also from mailed questionnaires, with a latency of 5 years from diagnosis. The mailed questionnaire was answered by 171 cases (95%) and 325 controls (90%). In 132 cases and 251 controls data on medical records were collected and compared with data from mailed questionnaires for the same subjects. No earlier X-ray investigation based upon medical records was reported in 39 cases and in 74 controls. The median cumulative thyroid dose for cases was 1.1 mGy in medical records and in mailed questionnaire 0.6 mGy. For the controls the respective median cumulative dose was 1.0 mGy in medical records and 0.3 mGy in mailed questionnaires. The median ratio between medical records and mailed questionnaires in the cases was 2.1. For the controls the median ratio was 2.9. Wilcoxon's Matched Pairs Test (WMPT) showed a significantly underreport of thyroid X-ray dose in both cases (P<0.01) and controls (P<0.000 001). In cases younger than 50 years at the time of the diagnosis of thyroid cancer there was not a significant underreport of X-ray examinations. Corresponding data from the control group showed however a significant underreport. Both cases and controls older than 50 reported significantly fewer X-ray investigations compared with data from medical records. A significant underreport was found among women both in cases and controls. For men there was a slight underreport among both cases and controls, although not significant. In conclusion, when studying diagnostic X-ray investigation as a risk factor for thyroid cancer, it is important to reduce the potential for recall bias when the study relies only on case–control reporting. To complement a case–control study with prospective medical data recorded at the time of the investigation could be an appropriate way to reduce the risk for recall bias.