OBJECTIVES: This study evaluates the quality (completeness and accuracy) of cause-of-death (COD) statements in infant death certificates as entered into a vital records system and as sesses its impact on infant mortality statistics in Hermosillo, Sonora, Mexico. METHODS: COD statements in a systematic random sample of 200 infant death certificates were compared to their corresponding medical charts. The underlying CODs (UCODs) orig inally recorded in each death certificate were contrasted with those assigned by an expert re viewer. Coding for the original and "new" UCODs was based on the three-digit category of the International Classification of Diseases, 10th Revision. Measurements of agreement be tween the two sets of UCODs were calculated and logistic regression was performed to deter mine factors associated with agreement. RESULTS: Overall agreement between the original and new UCODs was 52%. Agreement was excellent for the group of deaths due to congenital malformations, deformations, and chro mosomal abnormalities (kappa = 0.77); substantial for conditions originating in the perinatal period (kappa = 0.74); and poor for certain infectious and parasitic diseases, and respiratory diseases (kappa = 0.35). Overestimation (false-positive reporting) was highest (13%) for peri natal conditions, while underestimation (false-negative reporting) was highest (71%) for cer tain infectious and parasitic diseases, and respiratory diseases. Agreement was associated with type of UCOD (endogenous versus exogenous) and time of death. CONCLUSION: More than half (53%) of COD statements in infant death certificates in Her mosillo were inaccurately completed, which may lead to inaccurate interpretation of causes of infant mortality. Systematic assessments of the quality of COD statements may improve the quality of mortality statistics.