The objective of this article is to describe findings from a medication error (ME) survey, to estimate the extent of ME underreporting by comparison of survey results with written incident reports (IRs), and to determine factors associated with IR reporting of MEs. Participants were registered nurses from the 38-bed infant unit of a pediatric hospital. Most recent ME in each of four stages of the medication process was classified as to: timing, nature, whether the error was prevented from the patient, patient injury, and completed IR. Surveys were administered to nurses during mandatory skills session and were compared with IRs for MEs for the previous 6 months. The survey response rate was 93.5 percent; 72 nurses described 177 errors, 40.3 percent observed an ME in the previous week, 62.1 percent were prevented from reaching the patient and the likelihood of prevention was reduced in the later stages of the medication process. About 30 percent of MEs resulted in IRs. Administration errors were more likely to result in IRs compared with ordering errors, especially when the error was not prevented from the patient. There were 51 IRs for MEs. A multivariate logistic regression with completed IRs as the dependent variable showed a decreased likelihood of IRs for ordering than administration errors. IRs were more likely for wrong medication or dose errors and IRs were less likely for errors prevented from reaching the patient. The study found that by augmenting IR reporting of MEs and classifying errors by stage, anonymous ME surveys can be used for monitoring and guiding improvements to hospital medication systems.