Information on coronary heart disease (CHD) obtained from the Finnish Hospital Discharge and Cause-of-Death Registers was compared with that collected in the Helsinki Heart Study (HHS) during an 8.5-year follow-up. The purpose of the comparison was two-fold, firstly, to study the accuracy of registration of CHD and secondly, to find out what diagnostic codes to use for CHD in register-based follow-up studies. The HHS cases were used as the 'golden standard' and the CHD deaths and definite nonfatal acute myocardial infarctions (AMIs) (all diagnoses) were taken from the registers to establish the sensitivity of the Hospital Discharge and Cause-of-Death Registers combined. The sensitivity was 0.84 during the period 1980-86 and 0.87 during 1987-90, with the positive predictive values 0.94 and 0.92 respectively. The treatment effects seen in the HHS were compared with the effects that would have emerged, if register-based information only had been used with different definitions of CHD. Of the register-based calculations, the one with the definition 'all CHD deaths and hospitalizations with the ICD-8 code 410' came closest to the HHS result, with a 32% reduction (P=0.028 one-sided) of CHD incidence, while the original HHS result was a 34% reduction (P=0.008 one sided). However, when comparing Kaplan-Meier plots of cumulative hazards of CHD, the plot with a wider definition of CHD (ICD-8 and ICD-9 codes 410-414) came closest to the HHS experience, especially if revascularizations were included in the latter. Definite AMI as a single definition of CHD might thus not be sufficient when studying CHD risk, instead, at least two parallel definitions of CHD should be used.