The stability of the relationship between trends of violent deaths as a whole, suicides and death from undetermined causes was assessed by quantitative methods for the whole former USSR in the years 1970-1990 and for each of the 15 republics in the former USSR for the years 1984-1990. Semi-structured interviews during 1989-1996 were performed with 12 professionals involved in the diagnosis and coding of causes of death in the Baltic states and Russia. The quantitative analyses showed that mortality data were reliable for the Slavic (Russia, the Ukraine and Belarus) and Baltic (Estonia, Latvia and Lithuania) republics, and also for Kazakhstan, Kirgizia and Moldova. The Central Asian and Caucasian republics showed greater variation in trends and ratios, indicating a need for further investigations of the reliability of suicide statistics in these regions. It emerged from the interviews that no instructions to falsify data were given. The only instructions given were to treat the data on suicide and murder as 'top secret'. None the less, certain classification errors might arise in the statistics. The diagnosis 'undetermined cause of death' was permissible only as a preliminary diagnosis for 14 days, and there was a risk of criticism for poor-quality work if too many deaths were classified as being due to undetermined causes. Misclassifications could also occur in cases where there was a wish to conceal murder. Negative attitudes towards suicide, especially in Muslim regions, where suicide is taboo, might also have contributed to under-reporting of suicide.