According to the international literature, the proportion of homicides in infant mortality is certainly underestimated, and confusion exists during the certification process between homicides, sudden infant death syndrome (SIDS), and deaths "of unknown cause," with the percentage of the latter remaining stagnant in France. A retrospective study of suspicious infant deaths (1996–2000) involved hospital departments receiving deceased infants, district attorney’s offices, and statistics from the Center for Epidemiology on Medical Causes of Death (CépiDc) at Inserm, with cross-referencing of the three types of data (following approval by the National Commission on Informatics and Civil Liberties). The objectives explored two areas: 1) statistical data on violent infant mortality and the measurement of the magnitude of the problem; 2) professional practices leading to this epidemiological knowledge. The results concern 619 cases from 33 hospitals and 247 cases from 26 public prosecutor’s offices, in three regions: Brittany, Île-de-France, and Nord-Pas-de-Calais. In hospitals, the diagnosis of death suspected to be intentional or certainly violent and intentional (MSV) accounted for 5.3% of cases; when suspicion arose, a report was filed with the judicial authorities in one-third of cases, with very low feedback from the judicial system (13%). In the public prosecutor’s office survey, 80 MSV cases fall into three main categories: neonaticide (27), shaken baby syndrome (35), and “other” violent intentional deaths (18). Half of the children (excluding neonaticide cases) had a history of abuse, known prior to death in one-third of cases. In both investigations, under-investigation was observed, particularly regarding fundus examination and autopsy (half of the MSN cases were diagnosed without an autopsy). Cross-referencing with data from the CépiDc revealed an underestimation of homicides (3 to 15 times more numerous than the official figure, depending on the investigation); diagnoses of accidental deaths, deaths from unknown causes, and unclassified deaths often mask homicides. The low reliability of mortality statistics is largely due to insufficient collaboration between sectors (failure to transmit hospital and forensic institute data to Inserm). (R.A.)