According to official data from the Mortality Information System (SIM) of the Ministry of Health, the number of deaths caused by homicides in the State of Rio de Janeiro has declined between 2006 and 2009, from 7,099 to 5,064, which implies a decrease of 28.7% in the period. However, as we argue in this article, there are evidences that this variation is due to error on data classification. We conclude, first, that the number of fatal violent incidents with unknown cause in Rio de Janeiro inexplicably has increased since 2007, a fact that completely clashes with the national standard. Secondly, based on microdata of SIM, we analyzed the different type of victimization and we identify differences in patterns, in relation to socioeconomic characteristics of the victims, but also in relation to situational elements of the incident. Finally, we developed a multinomial logit model to statistically reclassify deaths with undetermined cause, to be homicides, suicides or fatal accidents. Our estimates indicated that the number of homicides in Rio de Janeiro, from 2006 to 2009 remained relatively stable, but the number of hidden homicides increased sharply during this period. We estimated there were 3,165 homicides not recorded in 2009, which represent 62.5% of the registred homicide.